APPENDIX A: LITERATURE REVIEW

A.1 Executive summary

A thorough review of available literature was conducted to ensure that the research team was fully informed of advancements in mobile LIDAR technology, techniques, and current and emerging applications.  Research documents were obtained from industry magazines and websites, technical reports, peer-reviewed journals, and conference presentations produced by leaders across the globe.

The literature review touches briefly on the basics of LIDAR technology followed by a more in depth description of current mobile LIDAR trends, including systems components and software.  This review also provides insights on current and emerging applications of mobile LIDAR for transportation agencies through industry projects and academic research. An overview of existing quality control procedures used to verify the accuracy of the collected data is presented.  A collection of case studies provides a clear description of the advantages of mobile LIDAR, including an increase in safety and efficiency.

The final portions of the review identify current challenges the industry is facing, the guidelines that currently exist, and what else is needed to streamline the adoption of mobile LIDAR by transportation agencies. Most existing guidelines for geospatial data are typically developed for digital terrain modeling using data from a generic source. They are generally focused primarily on elevation (vertical) error assessment, rather than 3D error assessment.

Unfortunately, many of these guidelines do not cover the specific challenges and concerns of LIDAR use.  Some have been developed for airborne LIDAR acquisition and processing.  However, these do not meet the needs of many transportation applications utilizing mobile LIDAR, creating a number of gaps that cannot be filled without an in-depth set of guidelines developed specifically for mobile LIDAR systems.  Evolving technology and limited experience with mobile LIDAR presents challenges for many organizations that can be overcome through the development of consistent, national guidelines.

From this review, there is a lot of discussion of “what” is being done in practice, but not a lot of “how” and “how well” it is being done.  A willingness to share information going forward will be important to the successful use of mobile LIDAR.

A.2 Scope of review

This literature review establishes a current state of the art related to mobile LIDAR technology and its use in transportation applications.  Several sources of information were analyzed, including:

• Industry publications and websites
• Technical reports
• Peer-reviewed journals
• Conference presentations
• Presentations by industry leaders

This review is meant for a wide audience of transportation personnel who may or may not be familiar with LIDAR technology.

The first sections of the literature review focus on the basics of LIDAR and mobile LIDAR Systems (MLS).  The next sections focus on both current and emerging applications of mobile LIDAR in transportation project planning, project development, construction, operations, maintenance, safety, research, asset management, and tourism.  Next, the review discusses data quality control and challenges with MLS.  Finally, the review discusses best practices, lessons learned and existing guidelines for MLS.

Because mobile LIDAR technology is new and rapidly evolving, limited information related to its use are available.  Much of this information is verbally disseminated rather than documented for a variety of reasons.  Further, most information sources do not provide sufficient detail needed to understand this emerging technology.

This review, in conjunction with a questionnaire (Appendix B) provides a baseline for development of national, performance-based guidelines to assist professionals in using mobile LIDAR for transportation applications.

A.3 Basics of LIDAR

Light detection and ranging (LIDAR) is an active (i.e. energy is emitted) method for remotely sensing distant objects.  It can be used to generate 3D models.  Coordinates of the reflected object are determined by the angle of the emitted pulse and the range to the object.  The range measurements are determined by one of two methods, (1) time-of-flight or (2) phase shift.  Time-of-flight scanners precisely record the time it takes for an emitted laser pulse to reflect off of remote objects and return to the scanner, while phase shift scanners emit a sinusoidally modulated laser pulse, and calculate distance using a phase shift principle.  This method can be used to more precisely calculate the distance over short intervals (typically up to 75m), consequently resulting in a higher level of positional accuracy and much faster data acquisition rate.  These benefits, however, come at the expense of limited range.  As such, time-of-flight systems (typical maximum ranges: 100 – 1,000 m; as high as 6,000 m) are generally more common for civil engineering and transportation applications.

Most time-of-flight systems are able to distinguish multiple returns from a single pulse, known as echoes, which provide useful information for filtering data.  For example, in the case of a forest (Figure A-1), part of the emitted laser beam may strike the top of the trees (first return), part may strike the branches (intermediate returns), and part may (hopefully) return from the ground (last return).  Phase shift systems generally do not have this capability.

Figure A-1:  Example illustrating concept of multiple returns from a single LIDAR pulse.

To distinguish each echo, the distance between them must be greater than half the pulse length (Vosselman and Maas, 2010).  For example, if the pulse width is 8ns, objects must be greater than 1.2m apart to be distinguished (assumed speed of light is 3 x 10⁸ m/s and refractive index is 1.0).  This can be calculated by:

(A-1)

The amplitude of returned echoes can be recorded, and are based primarily on the reflectance of the object returning the echo.  This amplitude of returned echo, called intensity, can be used to assist in distinguishing between different objects in the scan view.  Vosselman and Mass (2010) discuss how intensity values can be used to distinguish between objects at similar elevations, such as a manhole cover on a street, or painted street markings (Figure A-2).  Figure A-3 shows an example of an intensity shaded point cloud obtained from MLS for an intersection in Arizona.   Yang et al. (2012) describe a methodology to automatically extract pavement markings from mobile LIDAR point clouds by exploiting such intensity measurement information.

How intensely a laser pulse is returned to the scanner is determined by many factors such as range, angle of incidence, atmospheric conditions, and the material properties of the object being scanned.  Some of these factors are normalized so that a consistent intensity value can be obtained from the same object at different locations (Soudarissanane et al., 2011).  For example, objects closer to the scanner will have a more intense return than objects further away; this can be normalized so that range does not contribute to the difference in intensities.

Figure A-2:  Painted street markings and manhole cover can be better distinguished in the intensity return colored image on the left.  (Data from a static scan).

Figure A-3:  Grayscale, intensity-shaded image of mobile LIDAR data, showing painted lines and other features.  (Courtesy of DEA)

Scanning sensors can record returning echoes from a single pulse in one of two ways, discretely, and full-waveform (Figure A-4).  In the discrete mode, the scanning sensor records the returns as a binary result (yes, there is a return or no, there is not a return).  Full-waveform scanning sensors are able to record the entire backscattered waveform (Vosselman and Maas, 2010).

The return of the full-waveform allows for advanced determination of the peaks, which may indicate additional returns that were not recorded in the discrete analysis.  Further, material properties and geometry are generally better distinguished by a full-waveform scanner.  For example, scanning at an oblique plane (Figure A-5) will return a pulse width greater than the initial scanner pulse width; whereas a flat plane would return the same pulse width.

Figure A-4:  Discrete pulses vs. full-waveform returns.

Figure A-5:  Increase of pulse width on oblique surface.

Remote assessment using LIDAR (Duffell and Rudrum, 2005) can provide high speed data collection in areas with restricted access and/or safety concerns.   Particularly, use of MLS on transportation corridors can minimize roadway delays.  LIDAR sensors have been equipped on static ground-based platforms, and mobile platforms such as airplanes, vehicles (Figure A-6), boats, helicopters, UAVs, etc.  In “Stop and Go” scanning, a static scanner is mounted to a vehicle to reduce setup time.  The vehicle will periodically stop (e.g,, every 100 m) and perform a scan while the vehicle is stationary.  Much work has been done to develop and calibrate these devices for accurate surveying (e.g., Barber et al., 2008; Cahalane et al., 2010; Glennie 2007a, 2007b, 2009a, 2009b; Glennie and Lichti, 2010; Haala et al. 2008; Rieger et al., 2010).  The primary focus of this review pertains to mobile vehicular scanning, as opposed to airborne, railway, static terrestrial, and other platforms. Although airborne scanning has become more mainstream since the 1990’s (Duffell and Rudrum, 2005), often increased visibility, accuracy, and resolution needs require a ground-based scanning solution, particularly in transportation applications. Because static scanning has efficiency limitations, mobile scanning has become an effective solution to rapid data collection in recent years with advancements in scanning speed and accuracy, global positioning systems (GPS), and inertial measurement units (IMU).

Figure A-6:  Example of a MLS system (TITAN, courtesy of DEA).

A.4 MLS systems

A.4.1 Background and history

Prior to LIDAR based mobile mapping, other systems used a nearly identical platform setup but relied on photogrammetric methods.  The first fully functional system, GPSVan, was created in the early 1990’s by the Center for Mapping at Ohio State University.  It utilized GPS, gyro, DMI, two CCD cameras, and a voice recorder (Burtch, 2006).

Glennie (2009b) recounts the history of the first MLS system, constructed in 2003, which was a helicopter based LIDAR setup turned on its side and mounted onto a vehicle.  The system was used to survey Highway 1 in Afghanistan, which was potentially hostile for helicopter based scanning.  This initial system had many downfalls; primarily the limited field of view that accompanies airborne systems.  However, this system proved successful and demonstrated the potential value of MLS.  Currently, there are several MLS systems available through commercial vendors.  Yen et al. (2010) provides a comparison of many currently available mobile scan systems.

Mobile LIDAR systems provide a dense, geospatial dataset as a 3D virtual world that can be explored from a variety of viewpoints across a transportation agency.  With proper practices, this dataset can serve as a 3D model to link a variety of other data such as traffic data or crash data.

A.4.2 Components

Even though there are many MLS mapping systems, most systems consist of five distinct components:

1. the mobile platform
2. positioning hardware (e.g., GNSS, IMU)
3. 3D laser scanner(s)
4. photographic/video recording, and
5. computer and data storage.

A.4.2.1 Mobile platform

A mobile platform connects all data collection hardware into a single system.  The platform is usually a rigid platform, precisely calibrated to maintain the positional differences between the GPS, IMU, scanner(s), and imaging equipment.  It also provides a means to connect to the vehicle being used in the data collection process (Figure A-7).

Figure A-7: MLS system components (Topcon IP-S2 HD system operated by Oregon DOT).

A.4.2.2 Positioning hardware

Positioning hardware varies significantly from system to system.  However, at a minimum most systems incorporate at least one GPS/GNSS receiver and an inertial measurement unit (IMU).  The GPS/IMU system work together to continually report the best possible position.  In times of poor satellite coverage, the IMU manages the bulk of the positioning workload.  However, when satellite coverage is ideal, the IMU’s positional information is then updated from the GPS (Schwarz et al., 1993; Barber et al. 2008).  In addition to augmenting the GPS in periods of poor satellite coverage, the IMU must continually fill gaps between subsequent GPS observations.  Typical GPS receivers report positioning information at the rate of 1 to 10 Hz (i.e one to ten measurements per second).  However, during the course of a second, a vehicle will experience substantial movement, particularly when traveling at high speeds.  The IMU records positional information at a much higher rate, typically around 100 to 2,000 Hz, or 100 to 2,000 times per second (Shan and Toth, 2009; Yousif et al., 2010).  GPS/IMU data quality is typically the primary factor in gaining the best accuracy for a LIDAR point cloud (Ussyshkin and Boba, 2008).  Barber et al. (2008) explain how detailed route planning and satellite almanac checks can greatly improve accuracy with better satellite availability and geometry.

More complex MLS systems will utilize multiple GPS receivers, an IMU, and also a distance-measuring instrument (DMI) for improved positioning.  The DMI, a precise odometer, reports the distance traveled to improve GPS/IMU processing.  DMI’s provide direct distance traveled by measuring distance along the ground path, typically by mounting to one of the vehicles’ rear wheels.  In some MLS systems the DMI may be used only to trigger image capture at fixed distances (Kingston et al., 2006).

A.4.2.3 3D laser scanner

Many different types of 3D laser scanners are well suited for setup on a mobile platform.  These scanners are set to operate in a line scan (or planar) mode, where the scan head stays fixed and only internal mirror movement takes place.  Yoo et al. (2010) demonstrate how scanner orientation on the mobile platform can have drastic effects on the quality of data captured.  In order to minimize the number of passes necessary to fully capture data, most platforms utilize more than one scanner with view orientations at different angles.

A.4.2.4 Photographic/video recording

Photographic and video recording provides greater detail than the laser scanner alone (Toth, 2009).  The primary reason for this equipment is to color individual scan points in the point cloud to the representative real-world color.  This is done by mapping red, green, and blue (RGB) values to the geo-referenced point location.  This point coloring can make a highly dense point cloud appear as if it were a photograph.  Also, a visual record provided by this equipment can assist users in determining abnormalities in the scan data.  This imagery can be used by itself as a video log without the scan data, if needed.  McCarthy et al. (2008) discuss advantages to using combined LIDAR and photographic information for transportation applications including improved measurements, classifications, workflows, quality control checks, and usefulness. The scan data was particularly important for measurements on large objects such as bridges and embankments, while the photographs were most helpful for smaller objects.

A.4.2.5 Computer and data storage

Advancements in computer processing speed and data storage capabilities have lowered the cost, and increased the efficiencies of working with LIDAR data (Vosselman and Maas, 2010).  Mobile systems need to be capable of processing and storing large quantities of data from many sources.  The data includes: the point cloud, IMU, GPS , DMI , and all photographic and video data which must then all be integrated with a common, precise time stamp.  While some processing capabilities are available in the mobile system itself, much of the processing is still completed in the office.

A.4.3 System calibration

Accurate location of a ground coordinate from a mobile laser scan requires finding the value of 14 (or more, depending on the number of scanners) parameters for single scanner systems, each with a certain level of uncertainty.  These parameters are the X, Y, Z location of the GPS antenna, the roll, pitch, and yaw angles of the mobile platform, the three boresight angles from each individual scanner, the X, Y, Z lever arm offsets to the IMU origin from each scanner, and the scanner scan angle and range measurement (Glennie, 2007b).

Various methods can be used to help pare down some of the uncertainty of the individual values.  Barber et al. (2008) discuss a calibration procedure used to determine lever arm offsets, which consists of multiple passes over the same section of roadway.  The lever arm offsets will be propagated thorough the data set, and can be reduced by analyzing differences between the separate passes.

Boresight errors can also be determined by performing multiple passes over a region.  Glennie (2007b) discusses how these boresight values can be determined using a least squares adjustment to align the overlapping point clouds.   Rieger et al. (2010) also describe how boresight alignment of 2D laser scanners on a mobile platform can be determined by comparing to a reference 3D point cloud of the same region as well as a method of using multiple passes of an area to determine lever arm offsets between the IMU and measurement axis of the scanner.

Note that a system calibration should not be confused with a geometric correction or adjustment (sometimes called a site calibration).  A system calibration is done to correct for manufacturing errors and is typically done by the manufacturer.   This produces a set of parameters that remain constant as long as the hardware is not modified. (Although due to vibrations, with time systems need to be re-calibrated). A geometric correction or adjustment is done to correct for errors in the GNSS and IMU positioning information by adjusting the scan data to control or between adjacent passes.  This correction would be applied uniquely for each project.

A.4.4 Software and data processing

The scanner data consists of ranges, angles, and timestamps collected by the scanner, that are referenced from the scanner origin.  These measurements are then converted to XYZ coordinates as a point cloud (Figure A-8) when combining other sensor data (GNSS and IMU).  For most uses of MLS data, several processing tasks need to be completed:

1. Geo-referencing the data,
2. Mapping color information,
3. Filtering\cleaning of points, and
4. Generating models or extracting features from the point cloud.

Managing the process of acquiring data via an MLS survey requires extensive knowledge and experience.  Figure A-9 presents a typical workflow for MLS acquisition and processing, highlighting the key steps.  However, note that additional steps and procedures can be required depending on the applications of interest and end user data needs.  Also, data often must be processed using several software packages (both commercial off the shelf, COTS, and custom service provider) in order to produce the final products.  Finally, several stages will require temporary data transfer and backup, which can require a substantial amount of time (hours to days) due to the sheer volume of data.  Aside from geo-referencing the data, most processing tasks are similar between airborne, static TLS, and MLS systems.

Figure A-8:  Point cloud data of downtown Santa Ana, CA obtained through MLS (Courtesy of DEA).

Figure A-9:  Generalized MLS workflow, including interim datasets.

A.4.4.1 Geo-referencing

A prime interest in software processing is to register, or combine, many independent 3D point clouds into a single data set referenced in a single coordinate system with minimized error (Brenner, 2009).  Point cloud data must undergo several software processing procedures to accurately position the point cloud in the selected coordinate system.  Components of the MLS system simultaneously collect and store data (e.g., the GPS stores location, the scanner collects point locations relative to its origin, the IMU provides location corrections, and the color information is collected by photographic or video methods).  This data must be precisely time-stamped for integration (Rieger et al., 2010).  RTK GPS or post processed kinematic (PPK) GPS are the primary methods employed to geo-reference the MLS data; however, other methods (Barber et al., 2008) can be utilized such as alignment to targets, high resolution TLS data, or ground control points surveyed through traditional methods.

Often, alignment to high resolution TLS data and/or ground control points are used as a post-processing validation step to provide a measure of how accurately the MLS system has performed.  In areas where the GPS/IMU system did not collect accurate geo-referencing data, the MLS point cloud may be adjusted to ground control through a least squares adjustment.  Adjustments (Geometric corrections) are often implemented between passes to correct for biases.  Data processing can also introduce additional errors into a point cloud, but generally it will bring a point cloud into a much higher level of accuracy than the originally captured point cloud, depending on the applied processing procedures (Ussyshkin and Boba, 2008).

A.4.4.2 Mapping color information

As a LIDAR scanner collects data, a precisely calibrated image recording system can collect color information to map to each individual point in the point cloud (Vosselman and Maas, 2010).  This color information is stored as a numerical value (e.g., 0-255) in the red, green, and blue spectrum (RGB).  This color mapping is typically tagged to the individual points in a point cloud so that a location given as X, Y, Z is then amended to include R, G, B values (i.e. X,Y,Z,R,G,B).  In some instances, calibrated images can be overlaid on a point cloud adding X, Y, Z data to a 2D image.  This provides users more accustomed to working in a 2D environment the ability to transform 2D drafting into a 3D environment (Knaak, 2010).

A.4.4.3 Filtering of points

Following registration, point cloud data is typically filtered to eliminate unwanted features, including pits and birds, objects passing in the scanner view, unwanted vegetation, or, more generally, anything that is not needed by the end user.  Filtering is also commonly done to reduce the file size of the deliverable point cloud since the full dataset can require intense computational power and data storage.  Some common filtering techniques include:  first, intermediate, and last returns, selection of every i^th point, minimum separation between points, spatial hierarchy (e.g, octree or k-d tree), elevation, range, and intensity (see Vosselman and Maas, 2010, for examples of filtering algorithms).  Note that octree and k-d tree structures are also generally used as data organization schemes to improve interactivity of the dataset.

A.4.4.4 Generating models from the point cloud

Mathematical computations are not easily performed on point cloud data.  Typically, these point clouds are modeled using triangulation or gridding techniques for bare earth models, or by applying least square fitting of geometric primitive shapes (e.g., planes, squares, rectangles, cylinders, or spheres) to the structures found in the point cloud.  Typically, modeling of features in a point cloud incorporate an automated or semi-automated segmentation algorithm; this algorithm predicts points that can be modeled to a real world object, permitting extraction of the modeled structure (Vosselman and Maas, 2010).   More discussion of feature extraction will be presented in section A.6.9.  Various calculations and analyses can then be applied to these models to permit complex calculations such as volume change (e.g., Olsen et al., 2009).

A.4.4.5 Software considerations

In general, the requirements for software packages used for analyzing MLS datasets vary with respect to the final application of the dataset, and the variety of sensor data collected during the survey.  However, as a baseline, Rieger et al. (2010) describe four tasks that should be possible in various point cloud software programs:

1.  All data should be organized into one project where it can be processed and archived.
2. The data should be viewable on different scales, such as micro-scale point clouds and a full project area (e.g., as a rasterized data set).
3. The software should allow for geometric correction of the various sensors via a strip adjustment.
4. The data should be able to be exported in many different formats, including standardized formats such as LAS and E57, to be compatible with other software.

A.4.5 Scan deliverables

Common deliverables following laser scan projects include point clouds, CAD models and DTMs.  The options, advantages, and disadvantages of each deliverable type can be confusing for someone without substantial laser scanning experience.  Guidelines for accuracy reporting have been developed by ASPRS (2005) for airborne LIDAR, and many commonalities can be associated to MLS.

Providing adequate metadata on employed processing and filtering methods can be a challenge.  Additionally, because the technology and hardware evolve rapidly, it is difficult for software development to keep pace.  In conventional surveying, a point is tagged with a code for later identification during acquisition.  In mobile scanning, however, the collected points no longer are individually tagged with specific reference information; additional reference information must be added to individual points through semi-automatic or manual methods.

A.4.5.1 Metadata and specifications

There is currently no standard for reporting instrument specifications (e.g., POB, 2010 lists specifications for current systems, but varying techniques are used to determine the specifications) for static and kinematic laser scan systems, leading to potential confusion when comparing models and systems.  Additionally, because the specifications are developed in carefully controlled laboratory testing, they can create unrealistic expectations for data acquired in the real world, which varies significantly based on the application and materials to be scanned.  For example, some scanners are better suited for short vs. long-range applications and topographic vs. metal surfaces.  Many factors influence overall accuracies and resolution including:  range from the vehicle, objects blocking view, material, and speed of the vehicle.  The ASTM E57.02 subcommittee is currently working on developing standardized test methods for medium-range 3D imaging systems.  Glennie (2007b) recommends that at a minimum a boresight calibration report, and any confidence statistics should be included in the standard deliverables for a survey.

A.5 Mobile scanning advantages

A.5.1 Safety

Yen et al. (2011) show that MLS technology presents multiple benefits to transportation agencies, including safety, efficiency, accuracy, technical, and cost.  Mobile mapping has increased safety benefits over traditional survey techniques and static TLS (Glennie, 2009b), including safety and logistic improvements because nearly all work is performed from within the vehicle.  There are various reasons why this is beneficial:

1) Drivers become distracted by survey instruments, often observing the equipment and not paying attention to the actual surveyor.

2) Traffic often needs to be stopped or re-routed to allow the surveyor to make the necessary measurements.

3) Surveyors may have no other option but to place themselves in precarious situations to acquire the necessary measurements, whereas mobile mapping requires little or no need for surveyor and vehicular interaction.

4)  The vehicle generally can move with the flow of traffic, eliminating the need to divert traffic or close roadways.

A.5.2 Efficiency

Glennie (2009b) provides an example of MLS efficiency over a four mile section of a busy interstate section.  Washington DOT specifically requested that the roadway remain fully open for the duration of the survey, leaving MLS as the logical data collection method; total scanning time was 1.5 hours.  Mendenhall (2011) gives details about the cost and time savings of performing a MLS in San Francisco over 15 miles of roadway from the Golden Gate Bridge to the Palace of Fine Arts.  The cost saving on this project was estimated at $200,000 to $300,000 while the physical survey time was reduced by six to eight weeks further reducing management time by four weeks.

A.5.3 Comparison with airborne systems

Airborne and MLS share a number of similarities in the data processing workflow as both systems require the processing of positional data (e.g., GNSS, IMU) in tandem with LIDAR data.  Per mission, airborne LIDAR can be significantly more costly than MLS if solely focused on highway corridors, and does not provide the same level of detail from the ground plane.  On demand data capture can be provided by MLS, as well as capture of building facades and tunnels that are not available from airborne LIDAR (Barber et al., 2008 and Haala et al., 2008).  However, airborne systems can cover larger portions of the terrain and are not limited to ground navigable terrain.

Key differences between mobile LIDAR (MLS) and airborne LIDAR (ALS) systems (A-10) include:

• Airborne scanning is performed looking down on the ground. Given the larger altitude of flight compared to terrain elevation variations (except for steep mountains) and limited swath width, point density tends to be more uniform than mobile LIDAR. Mobile LIDAR systems will collect data more densely close to the scanner path and less dense farther from the scanner path.
• The laser footprint on the ground is normally much larger for airborne LIDAR than for mobile or helicopter LIDAR. This leads to more horizontal positioning uncertainty with airborne LIDAR.
• ALS generally will have a better (more orthogonal) view (i.e., look angle) of gently sloping or flat terrain (e.g., the pavement surface) compared to that of a mobile LIDAR system (depending on how the mobile laser scanner is oriented). This means that MLS systems will likely miss bottoms of steep ditches that cannot be seen from the roadway.  However, mobile LIDAR systems will have a better view of steep terrain and sides of structures (e.g., Mechanically Stabilized Earth (MSE) walls, cliff slopes). Jersey barrier will block line of sight and create data gaps on the opposing side.  Some projects may benefit from integrated mobile, static, and airborne data collection.
• MLS can capture surfaces underneath bridges and in tunnels.
• MLS is limited in collecting data within a short range (typically 100 m) of navigable roadways. Airborne platforms have more flexibility of where they can collect data.
• For MLS projects, accuracy requirements are the most significant factor relating to project cost. For ALS, acquisition costs generally control the overall project cost.
• For MLS, the GNSS measurements are the major error source; whereas for ALS the IMU and laser foot print are the major error sources (except for low-flying helicopter LIDAR).

Similarities between the systems include:

• Both acquire data kinematically using similar hardware components (GNSS, IMU, and LIDAR).
• Both capture a point cloud.
• Both systems typically provide laser return intensity (return signal strength) information for each laser return.
• Each point is individually geo-referenced with both systems.
• While MLS can offer significantly improved horizontal accuracy due to look angle, both systems can provide data with high vertical accuracy.
• Both systems can simultaneously acquire imagery and scan data.

Figure A-10: Comparison of Airborne and mobile LIDAR systems.

A.5.4 Comparison with static scanning

Zampa and Conforti (2009) provide data showing that MLS can be significantly more efficient than static TLS.  For example, in 2007 a 80 km stretc.h of highway was scanned using TLS, and in 2008 60 km of similar highway was scanned using MLS.  The field time required to collect the TLS was 120 working days, while the MLS was able to capture all the data in three hours.

Static scanning can provide some advantages over MLS, especially flexibility.  Static scanning provides more options for setup locations, including away from the road.  Users can also determine the desired resolution at the single setup.  This enables static scanning to obtain higher resolution on objects such as targets.  Generally, higher accuracies and resolutions can be achieved since the platform is not moving.

A.5.5 Overall Comparison

Based on findings from a literature review and questionnaire, Chang et al. (2012)  provide a chart to aid in selection of platforms for several applications with a discussion of generalized comparisons between mobile, airborne, and static terrestrial platforms based on several criteria:

1.  Applicability – Mobile systems can provide survey/engineering quality data over faster than static scanning.  Airborne systems (with the exception of low-flying helicopter) generally do not provide survey/engineering qualtiy data.
2. Cost-effectiveness – Despite a higher initial cost than static scanning, MLS received a higher cost-effective rating due to long-term benefits of reduced acquisition time.
3. Data collection productivity – Mobile and airborne LIDAR were both more productive than static scanning.
4. Ease-of use – Because of the integration of multiple sensors and calibration of these sensors, MLS requires more training than static scanning.  However, it requires less training than airborne because a pilot is not needed.
5. Level of detail – static scanning provided the highest level of detail.
6. Post-processing efficiency – Airborne LIDAR had the best rating for post-processing efficiency and both static and mobile were given low ratings.
7. Safety – All platforms provided safety benefits; however, airborne received the highest rating due to limited traffic exposure.

A.6 Applications

MLS systems have been utilized along navigable corridors for a variety of applications including earthwork quantities, slope stability, infrastructure analysis and inventory, pavement analysis, urban modeling, and railways (e.g., Grafe, 2008).  Ussyshkin (2009) presents additional potential applications of MLS derived from existing airborne applications, such as topography, utility transmission corridors, coastal erosion (e.g., Olsen et al., 2009), flood risk mapping, watershed analysis, etc.  Duffell and Rudrum (2005) discuss additional applications of ALS, which are applicable to MLS, such as feasibility studies, route alignment, environmental assessments, 3D visualizations, noise assessment, vegetation management planning, and accident investigation.  Chang et al. (2012) provide individual summaries for a variety of applications of LIDAR usage (airborne, static, and mobile) for transportation applications.  The report also presents results from a questionnaire to state DOTs as well as internal discussions within NC DOT to identify these applications and document lessons learned.

CTC & Associates (2010) and Olsen et al. (2012) discuss general applications of LIDAR from various platforms in transportation.  In addition, the following applications demonstrate some more specific uses of MLS and the types of vehicles that these systems have been employed on.  These applications are far from exhaustive, especially as new applications of MLS systems are being realized on a frequent basis.  Figure A-11 provides a graphical representation of many of the discussed applications.

Figure A-11:  Graphical representation of common applications of MLS

The following subsections focus on both current and emerging applications of mobile LIDAR in transportation categorized by project planning, project development, construction, operations, maintenance, safety, research, asset management, and tourism.

A.6.1 Project planning

A.6.1.1 Roadway analysis

Grafe (2008) provides examples of a roadway digital surface model, cross sections, and a highway interchange that have all been surveyed using MLS.  Additionally, Grafe (2008) demonstrates how a controlled and guided roadway milling machine can be set to automatically cut the road using the digital surface model.  Olsen et al., (2012) show an example of how a vehicular model derived from a static scan can be used to evaluate its ability to navigate through a highway system that has been digitally captured through MLS, prior to travel.

A.6.1.2 Topographic mapping/DTM

As in ALS and TLS, topographic mapping is an important application of MLS, including earthwork computations.  Jaselskis et al. (2003) performed a comparative study of total station and LIDAR based volume calculations from TLS.  In this study, a 1.2 percent difference was calculated between the different methods, demonstrating that LIDAR can be a very efficient method of volumetric determination.

Vaaja et al. (2011) researched the feasibility of using MLS to monitor topography and elevation changes along river corridors.  The vehicles used in this study were a small, rigid hull, inflatable boat, and a handcart designed to be pulled along by an individual.  Results showed that MLS provides accurate and precise change detection over the course of the study (one year), however, very careful control of systematic errors need to be accounted for.  Vaaja et al. (2011) note that the scanning field of view was often parallel to the topography, resulting in lower accuracy than scanning conducted more perpendicularly to the topography.

Yen et al. (2010) evaluate the quality of DTMs of pavement created from MLS data.  They determined that although the technology does not currently meet Caltrans specification requirements, additional refinement of the technology should overcome this limitation in the near future.

A.6.1.3 General measurements

MLS systems provide a permanent record of site conditions that can be measured at any time after the initial collection of point data.  This allows users to remotely measure length, volume, elevation, deflection, smoothness, camber, curvature, and others (Jaselskis et al., 2005).  Figure A-12 demonstrates how linear measurements in a point cloud can be used to find lane width, sidewalk width, and building dimensions.

Figure A-12:  Linear measurements and point coordinates in a point cloud (static scan).

A.6.2 Project development

A.6.2.1 Development of CAD models for baseline data

Mobile LIDAR data are often converted to CAD models to serve as baseline information.  Much work is still manual; however, automated algorithms are continually being implemented and refined.  Section A.6.9 Asset Management will discuss more details about feature extraction and implementation.

Jacobs (2005) provides many examples of how baseline data can be used for further construction development; these include: slope stability near the roadway, intersection improvement projects, pavement quality monitoring, pavement volume calculation, roadway milling settings, and pre-accident condition data.  Figure A-13 shows MLS data used for planning purposes the Columbia River Crossing Project between Oregon and Washington.  MLS data were acquired on several arterial roads for baseline, geometric data for both planning and design.

MLS was used by the NC DOT to survey five sections of interstate highway (Mabey, 2009) to generate baseline drawings for design.  The MLS data met the engineering specifications and the acquisition was completed in 9 days compared to the estimated 50+ days that would have been required using fixed terrestrial laser scanning.

Figure A-13:  Plan view of a section of MLS data obtained for several arterial roads for the Columbia River Crossing Project, a comprehensive industrial, residential, and infrastructure redesign centered on the I-5 Bridge crossing the river.  (Courtesy of DEA).

A.6.2.2 Virtual, 3D design of alternatives

A LIDAR point cloud allows designers to test various configurations in a virtual world that recreates the real world in high accuracy.  The University of Wisconsin—Madison has utilized MLS to create a virtual world of the roadways surrounding the campus which is used in their driving simulator, allowing the simulator’s users to intimately connect the simulated environment with the real world (Mandli Communications, 2011).

A.6.2.3 Clash detection

MLS systems are capable of providing clearance data (Figures A-14 and A-15) for highway overpasses, bridges, traffic signs, and even roadside high power lines.  In many of these instances the network (absolute) geo-referencing accuracy of the point cloud is less important than the relative accuracy provided by the scanner (Clancy, 2011).  Olsen et al., (2012) provide examples of bridge height clearances over roadways and waterways for Oregon DOT.  These height clearances can be used to determine if a modeled object can navigate safely through the constricted section.

Vasquez (2012) describes a high publicity example of using a MLS point cloud for evaluating obstructions along the 15 mile route taken by the space shuttle Endeavour to the California Science Center in Los Angeles, California.  Clash detection using a 3D model of the shuttle and the MLS data indicated over 700 clashes (155 were overhead lines).  Because of pre-identification of these clashes, conflicts were resolved ahead of time, enabling an efficient move with minimal interruption.  For example, utility companies were able to plan ahead and interrupt service for a minimal amount of time (within 1 hour) during the shuttle move.  The results were visually communicated through 3D visualizations and 2D cross sections.

Whitfield (2012) discusses the development of automated bridge clearance software that is being used by Caltrans to document bridge clearances for 7,250 bridges.  The clearances needed to be determined within 1” vertically and 3” horizontally.  It is estimated that there will be more than 100,000 measurements for these bridges.  The final point cloud is estimated to be 531 terabytes in size with an additional 28 terabytes of imagery.  Finally, the automation is estimated to have saved 1.2 million manual mouse clicks.  In a comparison to traditional techniques, MLS showed superiority in speed of acquisition and removed the difficulty in trying to manually find the points of minimum clearance.

Figure A-14: Clearance values measured perpendicular to roadway surfaceusing a static scan point cloud (Courtesy of Oregon DOT).

Figure A-15: Mobile LIDAR data of a section of the I-5 corridor in Sacramento, CA (Courtesy of DEA).

A.6.3 Construction

A.6.3.1 Machine control and construction automation

Singh (2008) discusses the role of laser scanning in machine automation for transportation applications, and how this use enhances efficiency.  Rybka (2011) demonstrates an entirely digital site planning project.  Periodic scans with a MLS permit initial design, estimates of percent completion, project compliance, and as-builts at project completion.  Rybka (2011) also discusses “Design to Dozer” a demonstration of construction automation hosted by Oregon DOT and the PPI Group depicting how MLS data can be used to create a DTM for machine control and construction automation to grade a site without ever having to drive grade stakes.  All grading is done entirely through equipment guided by GPS and a base model created from the 3D point cloud.  This presents an opportunity for cost savings, time savings, and improves site safety although no-actual job studies or cost comparisons are currently available.

A.6.3.2 As-built surveys

Singh (2008) discusses the role of a living survey database through all stages of the infrastructure life cycle through planning, design, construction, and maintenance.  In addition, digital, as-built records provided by LIDAR can provide significantly more detail than traditional methods (Su et al., 2006).  These digital records are particularly effective compared to traditional red lines on paper drawings.

A.6.3.3 Post construction quality control

In addition to providing high accuracy as-built records, MLS can provide quality control on the construction process.  Tang et al. (2011) discuss the use of algorithms for determining the flatness of concrete providing permanently documented results of the flatness defects, and permits users to remotely access the surface.  Kim et al. (2008) verify super-elevation slope values, curb design, and soundproofing wall design by creating cross sections of a roadway at 5m intervals.  The MLS data can then be compared to the original CAD drawings to ensure construction was completed within tolerance.

A.6.4 Operations

A.6.4.1 Traffic congestion

Traffic congestion typically results from human error, and automakers are researching methods to remove much of the human component from driving.  BMW has been developing a system called Traffic Jam Assistant to take over driving tasks when vehicle speed is lower than 25 mph.  The system relies on GPS and LIDAR along with other components to perform steering, braking, and acceleration (Barry, 2011).

Thorton et al. (2012) used mobile LIDAR to evaluate parking utilization along arterial roads at various times of the day.  They propose mounting MLS units on public vehicles such as buses, which could collect daily datasets along specific routes.  They also noted the potential for vehicle classification and parking duration from the repeat datasets.  Comparison of the automated approach to ground truth showed a small error rate of 1/340 vehicles.

A.6.5 Maintenance

Mobile LIDAR can also be used for maintenance purposes.  Many maintenance tasks are similar to those described in Section A.6.3, Construction.  Hence, the reader is referred to that section for more details.  One key advantage is that mobile LIDAR could enable a rapid As-Built, geospatial record of maintenance that was completed, reducing the need for future, repeat surveys (Singh, 2008).

A.6.5.1 Pavement analysis

The data collected for roadways can be used for several geometric analysis including stopping sight distances, adequate curve layouts, slope, super-elevation, drainage properties, lane width, and pavement wear.  For instance, Zhang and Frey (2005) found that road grade could be reliably determined (within 5% compared to design drawing data) with airborne LIDAR data.    Amadori (2011) found that mobile LIDAR can be an effective tool for cross slope determinations, particularly when identifying sections that are out of compliance.  Several pavement resurfacing vendors have found the data to be effective to reduce change orders and over-run costs for resurfacing projects.

Herr (2010) presents several examples of how MLS data can be used to evaluate pavement condition including rutting, ride quality, rehabilitation, texture, and automated distress.  He emphasizes that the acquisition of all of these data from a single, integrated point cloud represents a major paradigm shift for the industry where these data are acquired from a variety of sources.  Tsai and Li (2012) document controlled laboratory tests using laser profiling units to scan pavement at high detail at ambient lighting and low intensity contrast.  The system was effective in detecting cracks automatically; although scanner tilt angle, transverse profile spacing, and sampling frequency were key variables influencing the detection accuracy.

Chang et al. (2006) performed tests to compare the use of static 3D laser scanning, Multiple Laser Profiler (MLP), and rod and level surveys and found significant correlation (99%). As MLS accuracies increase, it may provide the ability to provide detailed surface roughness data, which are important to evaluate new pavement smoothness quality, resulting in significant incentives and disincentives for contractors.  Chin and Olsen (2011) have shown that static TLS data has potential for pavement smoothness evaluation, which determines significant financial incentives/disincentives for contractors on highway construction projects.  Potentially, scanner intensity information could be usable to determine the reflectivity of painted stripes, signs, and more.  (However, actual implementation requires continued research and development to appropriately normalize intensity values).  Scanner intensity information can also be used to highlight damaged sections of concrete (Figure 29) or asphalt pavement, which reflects light differently.

Figure A-16:  Intensity return used to highlight concrete cracking in a static scan (plan view).

A.6.6 Safety

State DOTs are required to submit Highway Performance Management System (HPMS) reports.  Many elements needed (e.g., road geometry) for this report can be acquired efficiently through a mobile LIDAR system (particularly when additional sensors are mounted to the vehicle).

AASHTO’s  Highway Safety Manual (HSM) includes algorithms that have been developed into SafetyAnalyst (network level) and the Interactive Highway Safety Design Model (IHSDM, project level).  Both of these input roadway data and provide safety evaluations such as expected crash rates.  Many of these inputs are geometric and can be captured with mobile LIDAR.

A.6.6.1 Extraction of features for safety analyses

Lato et al. (2009) demonstrate how rock fall hazards along transportation corridors can be monitored using MLS.  For this study, the monitoring took place from both railway and roadway based MLS systems.  In both situations, MLS provided increased efficiency and also the ability to monitor hazards in real-time.  The safety benefits from real-time monitoring also extend beyond locating unstable rock hazards.

A.6.6.2 Accident investigation

TLS systems have been used to document accident scenes, permitting the accidents to be moved off the roadway sooner, and allowing investigators to continue the investigation after all physical evidence has been removed from the scene.  3D Laser Mapping (2011) reports that accident scene investigation can be 50% faster than total station surveying, resulting in a 1.5 hour reduction in roadway closure.  According to Duffell and Rudrum (2005) and Mettenleiter et al. (2008), MLS has begun to play an important role in documenting pre-accident conditions, and also, a much faster means of documenting long accident scenes which typically occur in high speed crashes.  Jacobs (2005) discusses that laser scanning may also be used to analyze structural damage caused by vehicular impact on bridge overpasses due vehicle height exceeding the bridge clearance.

MLS systems can rapidly scan networks of tunnels for damage inspection.  Rapid deformation analysis enables highway crews to safely open a tunnel soon after a problem is resolved.  However, the resulting accuracy using MLS will depend heavily on the length of the tunnel and quality of the IMU because GNSS data will not be available in the tunnel.  Figure A-17 shows an example of an intensity shaded TLS dataset obtained for a tunnel damaged by fire.  Oregon DOT is planning to use their mobile LIDAR system to scan tunnels in Oregon on a repeat basis for monitoring.

Figure A-17:  Close examination of the intensity shaded point cloud (static scan) shows additional, minor damage to concrete in a tunnel in Oregon.  (Courtesy of Oregon DOT).

A.6.6.3 Driver assistance/autonomous navigation

Brenner (2009) and Toth (2009) discuss how MMS’s have begun to shape the research track of the autonomous vehicle navigation field.  Toth (2009) predicts that autonomous vehicle navigation could be operational within the next decade.  Brenner (2009) tests a simulated car, designed to model what a fully autonomous vehicle would be able to sense from a position on the roadway.  This is done by automatically extracting poles (any vertical narrow structure), and then allowing the autonomous vehicle to calculate positioning based on the constellation of the poles.  Pole extraction is performed on an already geo-referenced point cloud, and vehicle positioning calculated along the roadway based on referencing to the located poles.  Kodagoda et al. (2006) describe how laser systems on vehicles can be used to track curbs.

A.6.7 Research

A.6.7.1 Unstable slopes, landslide assessment

Su et al. (2006) describes the use of LIDAR data for geotechnical monitoring of excavations, particularly in urban areas.  In these urban excavations, real time monitoring of the excavation site as well as surrounding infrastructure is critical in maintaining integrity.  Miller et al. (2008) demonstrate the use of TLS in assessing the risk of slope instability, and provide two examples along transportation corridors.  The authors note the challenge and safety issues that arise from setting up a stationary TLS instrument along the side of a busy transportation corridor.  Figure A-18 demonstrates how LIDAR can be used to highlight localized slope failures.  Olsen et al. (2011b) developed an algorithm that permits in-situ detection of changes that have occurred over a region of previously collected LIDAR data using static LIDAR.  This allows field crews to immediately see where changes have taken place so that any additional measurements can be made at the site with no need for office processing of the point cloud.  Although mobile LIDAR data is not often processed in real time, it can provide baseline information for such a framework.

Figure A-18:  Static scan of a surficial slope failure along highway embankment at the US 20 Pioneer Mountain to Eddyville re-alignment project in Oregon.  (Note that the failure scarp is covered by a white tarp to prevent sediment from entering nearby water).

Lato et al. (2009)  found that mobile LIDAR was advantageous compared to static LIDAR in coverage, acquisition rate, and corridor operation integration.  Mobile LIDAR provided slope heights, angles, and profiles.  Using a rail mounted mobile LIDAR system, 20km of railway were acquired in 5 hours producing a 15 GB dataset with accuracies of 15 cm (absolute) and 5 cm (relative).  Figures A-19 and A-20 demonstrates similar use of LIDAR along unstable slopes for Oregon DOT and Alaska DOT to evaluate slope stability.

Although based on static scanning research, a pooled fund study conducted recently evaluated the use of LIDAR to map geotechnical conditions of unstable slopes, including rock mass characterization, surficial slope stability, rockfall analyses, and displacement monitoring.  The report (soon to be released) provides an overview of ground-based LIDAR and processing software, discusses how LIDAR can be integrated into geotechnical studies, and includes case studies in the states of Arizona, California, Colorado (two sites), New Hampshire, New York, Pennsylvania, Tennessee, and Texas. The authors also discuss best practices and procedures for data acquisition to ensure it provides reliable data for geotechnical analyses (Combs et al., 2012).

Figure A-19:  Point cloud of a rockfall on newly cut section for a highway. (Courtesy of Oregon DOT).

Figure A-20:  Point cloud for MLS data obtained for slope stability assessment on the Parks Highway near Denali National Park, Alaska.

A.6.7.2 Coastal erosion

Olsen et al. (2009) provide background on TLS (stop and go) of long coastal cliff sections.  TLS provides many advantages over traditional methods of monitoring coastal erosion, these advantages primarily coming from the density of the data points collected on the cliff faces.  This allows for in-depth monitoring of accretion and excretion along the cliffs, as well as monitoring of large land mass movements.  Figure A-21 shows an example of such change analyses using surface models derived from LIDAR data.  One of the challenges of working with TLS along these coastal sections is the necessity to time the ocean tides to prevent equipment and users from being submerged.  Young et al. (2010) compare ALS and TLS for quantifying sea cliff erosion.  The TLS data enables detection of finer-scale changes, however coverage is limited.  In many areas, MLS systems can rapidly obtain these finer-scale changes over a much larger region; this is important for coastal highways such as Highway 101 on the West Coast.

Figure A-21:  Time series change analysis for the Johnson Creek landslide along Highway 101 in Oregon obtained through “stop and go” scanning.  Orange indicates erosion and blue indicates accretion and seaward movement.

A.6.8 Tourism

Tourism is an emerging application of mobile LIDAR.  As tools to visualize point clouds from LIDAR systems become available, mobile LIDAR can provide a new generation of 3D, digital maps.  Kersten et al. (2009) describe the acquisition of mobile LIDAR in the historic peninsula of Istanbul.  Only 80 ha of the required 1500 ha were completed using static scanning in 6 months; whereas the remaining 1420 ha were completed in 3 months using mobile LIDAR.

A.6.9 Asset management

A.6.9.1 Inventory mapping

Duffell and Rudrum (2005) discuss inventory mapping as a secondary benefit that can be utilized from a point cloud.  Inventory mapping can include any structure, pavement, signage, traffic signaling devices, etc. that can be extracted from a point cloud.  Kingston et al. (2006) focus on both manual and automated feature extraction.  In addition to feature extraction, they also demonstrate the ability of software to automatically detect road signs and classify them by shape as defined by the Manual on Uniform Traffic Control Devices (MUTCD).

A.6.9.2 Modeling and inspection

Becker and Haala (2007) emphasize the need for detailed 3D modeling of urban landscapes for city planning.  They demonstrate an automated façade grammar building tool that can model building facades beyond the line-of-sight of the scanner by hypothesizing further facades based on the adjoining style.  Jochem et al. (2011) also proposes using MLS to model building facades; however, the focus is to select the facades with the highest solar potential.  The goal is to extract individual structures from a point cloud and assign solar potential ratings to the various facades of the structure.  This would allow individuals to easily see where the most appropriate placement for solar panels would be on their building.

A.6.9.3 Automated/semi-automated extraction of features

New algorithms are under development to extract features in a point cloud.  Many of these are currently semi-automatic and require significant user verification of results.  However, many researchers are developing robust, fully automated feature extraction tools.  For example, although primarily developed for robotics, the Point Cloud Library (PCL, http://pointclouds.org/) is a recent open source resource that has libraries for feature extraction from point clouds of geometric primitives (planes, cylinders, etc.).  Common features extracted from point cloud data include signs, streetlights\poles, reflective striping, and curbs.  Please note that many of these procedures currently have only been tested on limited, test datasets and have not been integrated into mainstream software.  However, current software is rapidly evolving to implement these novel techniques.

 McQuat (2011) discusses several different structures (signs, facades, bays, automobiles, curbs, et al.) including how they can be automatically detected and converted to useful shapes for use in a GIS.

Pu et al. (2011) describe automated algorithms to recognize features within a point cloud such as traffic signs, trees, building walls, and barriers using characteristics such as size, shape, orientation, and topological relationships to classify the point cloud.  The authors indicate that poles are recognized with an accuracy of 86%; however, other categories were not extracted as successfully and need to be integrated with imagery for extraction.

Semi-automatic or fully automatic extraction of signs is necessary to efficiently locate signs in a large point cloud such as that provided by MLS.  Figure A-22 provides an example of how the intensity values of the scanner can be used to identify reflective signage can be semi-automatically detected for extraction and cataloging.

Novak (2011) discusses the use of MLS to extract streetlights in El Paso, TX, and store them in a database managing light bulb replacement.  Due to an increase in worker safety and a faster rate of completion, MLS was chosen for the project.  Brenner (2009) discusses a method of pole extraction by use of cylindrical stacks; these stacks contain a core that must contain data surrounded by a ring that contains no data.  Lehtomaki et al. (2011) used MLS data to extract poles and trees.  The automated method successfully detected 70% of the poles and 78% of the trees at two field sites.  Of the detected features, 81% (poles) and 87%  (trees) were correctly identified.  The algorithm had difficulty recognizing tree trunks surrounded by branches and wall structures.

Rutzinger (2009) combine airborne and mobile LIDAR data to extract vertical walls for building facades.  These wall faces are then used to correct building outlines in cadastral map data.  Following point cloud segmentation through a region grow process, individual points are classified based on planarity, inclination, wall height and width.  Upon detection of a vertical wall, the MLS points are then compared to the vertical wall from the cadastral map to estimate the potential completion of the MLS data.  Vegetation, for example, created several occlusions.

Alabama DOT also recently implemented mobile LIDAR for maintaining a billboard inventory and found it to be a cost-effective system.

Lin and Hyypa (2010) developed an automatic methodology to detect pedestrian culverts from DTMs created from mobile LIDAR data.  Because of limited view of the culverts from the roadway, culverts could only partially be characterized.  However, calculated lengths and widths of the culverts were within 9% and 16% of actual measurements.

Figure A-22:  Reflective signs (red) extracted from a static TLS point cloudat the Oregon State University campus.

A.7 Data quality control

A.7.1 Accuracy and precision checks

Each component of the MLS setup requires careful calibration to ensure accurate data.  Calibration errors are additive in the scanning platform, each portion of the system that is not well calibrated propagates errors to the final point cloud.

A.7.1.1 Laser scanning errors

System specification sheets provide a basic idea of scanner performance; however, additional factors need to be considered that are well beyond the scope of the specification sheet.  Also, because standardized testing procedures have not been developed, it can be difficult to directly compare values from one system to another.  Error sources include the material properties of the scanned objects, environmental conditions, inconsistencies in scanner manufacturing, the geometric configuration of the object to the scanner, and GPS errors.

• Material Properties:  White surfaces will provide very intense laser returns, while black surfaces will return a much less intense value (Boehler et al., 2003).  System performance varies greatly, and consideration needs to be taken for the objects being scanned, such as low reflectivity asphalt in many transportation applications.  Highly reflective surfaces (e.g., traffic signs, retro-reflectors) may produce additional distortion effects such as saturation and blooming (Vosselman and Maas, 2010). Saturation (Figure A-22) is caused by too much energy being returned to the scanner and appears as points spread out along the line of sight of the scanner. Blooming (Figure A-23) is a similar effect that occurs perpendicular to the line of sight of the scanner, creating an apparent enlargement of the reflective surface due to excessive energy being returned to the scanner.

Figure A-23: Extreme case of Saturation of a flat, 5cm retro-reflective target (red) is seen as the target extending 4cm off of the wall.  Left: Straight on view (down on Y-axis).  Right: side view along plane of wall that target is affixed to (along X-axis).

Figure A-24:  Blooming (i.e., enlargement) of a flat 5cm retro-reflective target (red).

• Environmental Conditions:  Moisture in the air or on surfaces will often lead to data dropouts or noise in the data.  Heavy fogs will also limit data collection capabilities.
• Inconsistencies in scanner manufacturing:  Boehler et al. (2003) warn that many scanners are built in small quantities and individual errors vary significantly between units.  Careful care and inspection of equipment, in addition to periodic calibration checks are necessary to maintain the best possible accuracy from hardware.

Geometric configuration:  The size of the laser footprint is important in understanding the final data accuracy.  The uncertainty of point location due to divergence of the laser beam adds additional random error (Barber et al., 2008).  Boehler et al. (2003) state that it is possible to record the same object multiple times using multiple passes of the scanner, however, due to the beam width and angular uncertainty, the exact same point cannot be measured precisely.  The obliquity of how the laser pulse strikes the surface can result in significant positioning error (Laefer et al., 2009; Olsen et al., 2009; Olsen et al. 2011c).  If two objects\surfaces are placed less than half a pulse width apart, along the line of sight, a mixed pixel (Figure A-25) discrepancy may result (Vosselman and Maas, 2010).  This discrepancy can be seen as an extension of points off of the edge of the closer object, extending back to the further object. The return energy is split between the objects.

Figure A-25:  Mixed pixels appear in an area occluded from the scanners line-of-sight.

• GPS source errors:  Factors that affect the accuracy of GPS include:  multipath, shading by buildings and trees, loss of satellite lock, atmospheric conditions, and poor satellite geometry (Glennie, 2007b and Haala et al., 2008).  GNSS systems combining GPS, GLONASS, Galileo, and Compass (when available) will help improve accuracy results (Chiang et al., 2010).

A.7.2 Procedures for measurement quality control

Many different methods have been employed to verify the accuracy of the final point cloud.  Commonly, ground control points, or an already geo-referenced TLS point cloud are used to verify accuracy of the MLS data.  Ussyshkin (2009) discusses geo-referencing mobile scan data using a system of six base stations and ground control points spaced every 50-80 meters throughout the survey extents in order to achieve 1-2cm accuracy.  While this may be achievable for a small project, a MLS survey needed for a system-wide analysis could not be economically completed with this amount of control required.  Caltrans specifications call for these validation points every 500ft ( ~152m).  Barber et al., (2008) state automated validation to compare MLS data to survey control high resolution terrestrial laser scans, or target matching in real-time is greatly needed.  Hiremagalur et al., (2007) provide “best practices” to ensure the proper registration of MLS data and recommend target redundancy (if target registration is to be used), examination of overlapping point clouds, and comparison of point cloud coordinates to check point coordinates surveyed using traditional methods.  A report of the RMS error of the point cloud to the ground control coordinates should be a standard deliverable in addition to an RMS error report of overlapping point clouds.  Points to be used for an RMS evaluation should be spatially distributed throughout the entire dataset.  Additionally, Graham (2010) recommends that final quality control be performed by someone other than those involved in registering the dataset.

A.7.3 Data collection categories concept

When assessing the quality of a mobile mapping system point cloud, many factors contribute to the final accuracy and precision values.  Boehler et al. (2003) describe that various jobs will require various levels of data quality.  The ASPRS Mobile Mapping Committee (2011) and Hiremagalur et al. (2007) have recommended that final point cloud quality be assigned a rating based on the quality of data.  For example, an end user may be in need of a point cloud to inventory roadway signs along a corridor.  The user may not be concerned with the geo-referencing accuracy of these signs; they may be using the data solely for the purpose of counting the number of signs along the corridor.  In this example, the user would not want to pay a premium for survey quality positional data, which also requires additional field time to complete.  This user still needs high enough resolution in the point cloud to be able to reliably extract the signs.

This creates a two-fold level requirement for the data in that it needs to address both the accuracy and the resolution of the data (ASPRS Mobile Mapping Committee, 2011).  Accuracy tends to have a higher impact on project cost, since higher resolution can be more easily obtained with slower vehicle speeds, or multiple passes through the corridor.  According to Barber et al. (2008), positioning is not affected by vehicle speed; whereas, higher speeds lead to lower point density.

However, Duffell and Rudrum (2005) argue that the over-collection of data may not always be a negative, because data can often be reused for many different tasks.  One data cloud could be made available to many end users who can mine the data source for several different job tasks.  In addition, extra detail could allow the reuse of archived point clouds for base data in accident investigations, hazard identification, and future project planning.

A.8 Current challenges

Several difficulties exist when performing mobile scans (e.g., Glennie, 2009b). Measurements are performed from a moving platform, requiring high precision GPS/IMU readings for accurate data geo-referencing.  Typically it is not feasible to close down a section of highway for scanning, so neighboring vehicles can block data collection efforts.  Additionally, the vehicle must be moving at a safe speed (with the flow of traffic) while simultaneously collecting data. In some cases, a rolling slow down can be used to avoid these problems.

Further, the size and complexity of the laser scan data presents significant challenges.  Sensors collect data at very high speeds (typically 100k to 1 million points per second) and at very high point densities (typically >100 points per m²) at close ranges (typically < 100m). This creates large datasets that can be difficult to work with on typical computing platforms and software.  The volume of data collected also requires a substantial amount of data storage and backup during a project.

Following completion of a project, care must be taken to ensure proper data archival.  The large size also makes web, DVD, or other common media difficult to use for data transfer or sharing both within an agency and with external partners.  The complexity of data and minimal availability of software also presents challenges to end users such as transportation agencies in actually being able to use the data.  Ussyshkin (2009) discusses limitations on the number of points that can be imported into common software packages.  Currently, many consultants subsample and filter the data to reduce size.  They also process the data in small sections (tiles) because computing resources limit their ability to work with the entire dataset.  Often, the final data typically transferred to the end user may only represent a fraction of the original data obtained.  In several cases, the actual point cloud is not being delivered.

While manufacturers of GIS and CAD software have recently been integrating point cloud support, many challenges remain to make this process seamless for the end user.  Further, point cloud processing usually requires working between multiple software packages where information can be lost on imports and exports through the process. The ASTM E57.04 (2010) subcommittee on data interoperability was formed, in part, to help resolve these data transfer issues. In addition, working with a 3D point cloud requires skill to ensure that appropriate measurements are extracted.

Knaak (2012b), after a conversation with Florida DOT personnel, discusses problems with MLS technology adoption by transportation agencie and offers suggestions including:

1. Avoid the “WOW” factor of point clouds.  Often this results in incomplete projects where consultants do not provide transportation agencies with something they can actually use,
2. Agree on a QA/QC procedure, including a lineage from the point cloud to the final product and metrics to evaluate that lineage.  The QA/QC should be done by an independent contractor,
3. Identify the model needs first so that the point cloud requirements can be determined easier, and
4. Define the respective responsibilities of the customer and consultant in the process.

Knaak (2012b) also explains problems in current payment and procurement standards that are focused on time in field work and minimal office processing time.  The key factor with MLS technology is that it reduces field time dramatically (80-90%) but shifts loads to processing.  Under current payment schemes, these current payment schemes reduce the contractors pay substantially because they are paid based on field time.

A.9 Best practices and lesson learned

Unfortunately, many of the lessons learned, and user experiences are being disseminated verbally at conferences or other events, but currently have not been adequately integrated into retrievable documents. Many service providers are also reluctant to document and make project reports available because of liability concerns.

Missouri DOT (Vincent and Ecker, 2010) evaluated the accuracy, cost and feasibility of airborne, mobile, and static terrestrial laser scanning for typical transportation projects. They determined that all systems met their accuracy requirements. The report also highlights current hurdles including software and computing challenges. The authors also conclude that traditional surveying and/or static scanning may still be required to fill in gaps from mobile scanning.

Yen et al., (2011) provide an in-depth evaluation of MLS technology in the State of Washington.  They show that maintenance, asset management, engineering, and construction programs all incur cost savings, time savings, and safety improvements with MLS.  This evaluation also demonstrates the needs of national standards and best practices as well as a common data exchange platform to improve data interoperability.

Singh et al. (2012) present an overview of theory applied to mobile LIDAR and practical implementation for a case study of an 8 mile segment of the I-5 corridor.  This workshop presentation discusses project planning, quality management plans, data acquisition, data processing, deliverables and lessons learned.  Lessons learned include placing pre-marks (control points) on both sides of the run, providing significant overlap between cloud strips, breaking runs into manageable segments, planning for acquisition on lengths much larger than originally anticipated to cover frontage roads and ramps, and having flexible data storage and transfer mechanisms.

Many lessons learned have not yet been formally documented with rigorous testing results.  However, there are often “nuggets of wisdom” that can be found on various websites.  For example, many service providers and vendors publish short articles of projects and experiences on http:\\lidarnews.com.  Some service providers regularly update a blog, such as Michael Baker Jr.  Inc. (http://mobilelidar.blogspot.com).    As an example, this blog includes a discussion of “lessons learned” from their experiences.  Below is an excerpt:

 “Baker’s Dozen: 13 Laws of Mobile LIDAR (also currently being chiseled on a slab of granite):

1. Too much is better than not enough.
2. Sometimes more is just more, not better.
3. Hard drives are cheap, time isn’t.
4. Consistency counts; stop guessing.
5. When someone wants “full planimetrics,” they really don’t.
6. The stated laser range is X’, but the lasers are only capturing data to Y’; and Y is definitely less than X, yet nobody can tell you what Y is…
7. The data you capture is only as good as the applied control.
8. Today’s best practices will be tomorrow’s old habits.
9. Field vs. office time ratios are pipe dreams.
10. Mobile LIDAR systems are not created equal, and neither are the operations behind them.
11. Off-the-shelf processing software will only do 50% of what you need it to do.
12. When the system encounters issues, take a breath and reboot.
13. Mobile LIDAR is not all fun and games, but it does feel like it some days.”

Siebern (2012) presents two case studies and information on “managing expectations for mobile mapping solutions,” from the perspective of a service provider. Particularly, the author mentions that proper communication and understanding between service providers and clients is critical to project success, particularly related to the fact that the LIDAR industry is evolving,   The case studies (interstate corridor design and overhead catenary system) discuss various aspects of the projects including expectations, deliverables, challenges, and unforeseen benefits (e.g., usefulness of the imagery for other purposes than originally intended) associated with the projects.

Recently, Chang et al. (2012) through a questionnaire and literature review documented several important lessons learned for various transportation agencies, including:

1. Despite benefits of LIDAR, it is not a complete substitute for traditional surveying.
2. Due to technical difficulties with hardware and software, a trained technician is required for editing and extraction, which can be a costly investment to implement.
3. Specifications need to be clear, particularly with accuracy requirements regardless of whether it is in-house surveyors or third-party contractors.

Burns and Jones (2012) reported on the recent U-Plan project to collect mobile LIDAR data for all roads within the state managed by the DOT.  Key lessons learned include:

1. Ensure the DOT has the ability to store, distribute, analyze and utilize the data collected.
2. Build support from senior management
3. Prepare for potential lengthy procurement processes
4. Be prepared to work extensively with the vendor from selection to final data collection.  For example, they found that a weekly meeting with the data provider was beneficial to the project, and
5. Do not expect to fund your entire data wish list up front!

A.10 Existing guidelines

Many agencies (FAA, 2011; FGDC, 1998; NDEP, 2004; NOAA, 2009; USGS, 2012) have provided recommendations, guidelines, or standards for geospatial data.  Some of these (FGDC, 1998 and NDEP, 2004) are broad specifications that pertain to all remotely sensed data while others pertain more directly to LIDAR data (FAA, 2011; NOAA, 2009; USGS, 2010).  The ASPRS Standards Committee (2005) has produced “Guidelines Vertical Accuracy Reporting for LIDAR Data,” and “Guidelines Horizontal Accuracy Reporting for LIDAR Data” which more specifically declares reporting standards (e.g., fundamental vertical accuracy (FVA), consolidated vertical accuracy (CVA), supplemental vertical accuracy (SVA)).  A summary of these guidelines can be seen in Table A-1.

Common trends can be seen in the various LIDAR specifications, including:

1. Standard accuracy reporting methods,
2. Requirements for ground point density
3. Requirements for scan overlap,
4. Number and distribution of control/check points for accuracy verification, and
5. Types of deliverables.

Although most of these guidelines are currently focused on aspects of ALS, some of their fundamental principles can be adapted to produce guidelines more relevant to mobile LIDAR.  However, most of these documents do not directly or adequately address the needs of many transportation applications.  For example, the accuracy, resolution, coverage, and look angle of mobile LIDAR data varies significantly from that achieved with airborne LIDAR.   Particularly, true 3D error vectors are important for many applications that cannot be evaluated by focusing on vertical error only.

A.10.1 Geosptial Data Accuracy

The Federal Geographic Data Committee (1998) developed the National Standard for Spatial Data Accuracy (NSSDA), which provides guidance on reporting spatial data accuracies.  This document provides the foundation for the reporting found in most available standards and guidelines.   The NSSDA uses a root mean square error (RMSE) to estimate positional accuracy reported in ground distances at 95% confidence.  Datasets should be tested with a minimum of 20 control points and reported as:

Tested ____ (meters, feet) vertical (or horizontal) accuracy at 95% confidence level

In cases were the data were not tested and accuracy is merely estimated, the following statement is used:

Compiled to meet ____ (meters, feet) vertical (or horizontal) accuracy at 95% confidence level

The National Digital Elevation Plan (NDEP) guidelines further developed the NSSDA to include three types of accuracy reporting: fundamental vertical accuracy (FVA, open terrain, optimal conditions), consolidated vertical accuracy (CVA, combined accuracies obtained in all land covers), and supplemental vertical accuracy (SVA, accuracies reported for individual land covers).  For example, accuracies in dense forests will be much less than in open terrain.

Table A-1:  Summary of existing LIDAR guidelines.

A.10.2 ASPRS guidelines

The American Society of Photogrammetry and Remote Sensing (ASPRS) is striving to be the go-to source for LIDAR technology in the US.  Several efforts are underway, including:

• The ASPRS Mobile Mapping Committee is developing guidelines for mobile mapping.  This is currently a work in progress at the outline stage.
• ASPRS Vertical accuracy guidelines for airborne LIDAR.  This document reinforces the NSSDA and NDEP guidelines and provides guidance for establishing control specific to airborne LIDAR.
• ASPRS horizontal accuracy guidelines for airborne LIDAR.  This document provides background on the difficulties in determining horizontal accuracies from airborne LIDAR.
• ASPRS Geospatial Procurements (DRAFT).  This document is intended to aide entities with the best approach to commercial geospatial products, defined with a COTS specification.  The document distinguishes between professional\technical services and commercial geospatial products.  It also recognizes state and federal laws.  A proposed procurement methodology of license data terms and conditions, cost/value, service provider defined technical specification, services to support geospatial products and deliverables are addressed.   ASPRS also previously produced procurement guidelines for geospatial mapping services.

A.10.3 Transportation agency LIDAR standards

Chapter 15 of the California Department of Transportation (2011) Surveys Manual is one of the first developed set of specifications that explicitly addresses the required information and data quality that should be provided with static and mobile LIDAR surveys.  These specifications contain a two part classification system for mobile LIDAR surveys.  Type ‘A’ is a higher accuracy, hard surface survey used for engineering applications and forensic surveys.  Type ‘B’ is used for lower accuracy earthwork measurements (e.g., asset inventory, erosion, environmental and earthwork surveys).

These specifications are broad enough to not limit service provider equipment and technology but provide details regarding data acquisition and processing procedures, including the minimum overlap between scans, maximum PDOP, minimum number of satellites, maximum baseline, validation point accuracy requirement, IMU drift errors, and other factors pertaining to the geo-referencing accuracy of the point cloud.  However, one needs to have a relatively high level of understanding of mobile LIDAR technology in order to utilize these aspects of the Caltrans standards effectively.

Other transportation agencies have begun developing standards and guidelines for MLS.  These Guidelines are meant to provide the agency with a reference document that can be tailored to their specific needs.  For example, Florida DOT recently released guidelines which are very similar to the Caltrans guidelines.  However, the Florida DOT guidelines add a Type C, Lower Accuracy Mapping category for planning, transportation statistics, and general asset inventory surveys.

A.10.4 FAA Advisory Circular

The Federal Aviation Administration has produced a draft Advisory Circular related to remote sensing technologies.  This document includes a section which discusses considerations for use of several forms of LIDAR (static, mobile, and airborne) for airport surveys and anticipated accuracies and resolutions for each method.  The document also discusses calibration procedures for LIDAR systems and provides guidance when such calibrations are necessary.  Specific requirements for mobile LIDAR workflows include:  redundancy, monitoring acquisition, local transformation and validation points, data processing, data filtering and clean up, geo-referencing, and data integration.

A.10.5 Industry Guidelines

Some service providers have developed guidelines for transportation agencies that they have worked with.  Many of these are not published and can differ by transportation agency, to meet their individual needs.  For example, Knaak (2012a) has developed a set of best practices based on experience; this document defines three distinct levels of data as well as requirements for: vehicle trajectory, point cloud, file management, and images.

A.11 Motivation and key needs for national guidelines

Mobile LIDAR data provides many benefits when processed and used appropriately.  Ussyshkin (2009) states that the underlying technical details (e.g., applications, procedures, benefits) need to be well understood in order to prevent disappointments and misunderstandings when using mobile LIDAR data.  Guidelines need to incorporate and integrate fundamental principles of quality control and performance to result in the desired deliverable.  Optimally, end users such as engineers and designers should have a strong understanding of mobile LIDAR, so that the data can be utilized effectively and to its full potential.  However, because of the wide variety of applications and quality needs, many personnel within a transportation agency can effectively use mobile LIDAR without being experts in the details of the technology once the appropriate guidelines are in place.  Simple, yet powerful, guidelines focused on performance evaluation will enable them to adequately integrate mobile LIDAR into their operations.  National guidelines will ensure that transportation agencies do not duplicate efforts in producing similar documentation.  The consistency provided through national guidelines will also enable improved communication between service providers and transportation agencies.

A.12 Conclusions

This literature review highlights the use of mobile LIDAR in transportation, including a discussion of current and emerging applications, data quality control, existing guidelines, and challenges.  The review shows that there is a lot of interest for mobile LIDAR in transportation, provided appropriate guidance is in place.

From this review, there is a lot of discussion of WHAT is being done, but not a lot of HOW and HOW WELL it is being done.  Generally, most information related to MLS use are presentations at conferences or short web articles that do not go into detail regarding the work performed.  Most quality control checks that are discussed in these reports are verified for vertical accuracy only.  Very limited research exists to understand fully the capabilities and limitations of these systems.

Given the limited amount of experience that has been documented in the literature, to date it is important that future demonstration/pilot projects be adequately documented and the results disseminated both within a transportation agency and between agencies regarding the challenges, successes, and lessons learned from projects incorporating mobile LIDAR.

The literature review, in conjunction with the transportation agency questionnaire, reveals that there is a strong transportation agency\industry desire for:

• Standardized accuracy reporting methods
• Data interoperability and management
• Control/check requirements and procedures
• Better understanding the data quality needs of specific applications(e.g., asset management vs. engineering needs)

Another important consideration is that MLS is a tool in the transportation agency’s toolbox, sometimes it may be the best tool for a job, sometimes not. Hence, it is important that agency understand when to and not to use mobile LIDAR.