Airborne LiDAR—Rivaling Conventional Techniques
Recent advances in airborne LiDAR performance, processing software and positioning technology shows that a high precision LiDAR sensor is capable of producing data that rivals the precision and accuracy of conventional ground surveying techniques, and at a fraction of the cost and time.
By William T. Derry, PLS, M.SAME
LiDAR Point Cloud Data of airport runway. IMAGES COURTESY AXIS GEOSPATIAL.
Airborne LiDAR has been available for decades, and has undergone a remarkable change in its capabilities. Many of the advances in this complex technology are a result of changes in the individual components of the LiDAR sensor system, but they also can be linked to the technology used to process and prepare the data for the end user. Acceptance of the technology and its data has driven manufacturers to improve their systems and software. This, in turn, has allowed data providers to develop new techniques and find new applications for the data.
It is important to understand that not all airborne LiDAR datasets are created equally. The widespread availability of point clouds on the internet has caused some problems when that data is used for a different application than what was originally intended. Many LiDAR datasets are acquired for wide-area mapping projects, watershed mapping, forest cover studies or other non-engineering uses. This is done from substantial altitudes, often covering hundreds or even thousands of square miles, using sensors with precision specifications measured in feet, not fractions of a foot.
Review of the metadata file for a dataset can quickly show the user the technical aspects of the data they are hoping to use, so it is critical to understand the implications of the information it contains. The metadata that accompanied the dataset identified the sensor employed. Its spec sheet revealed that its precision was only one decimeter per kilometer of range measured. The acquisition altitude and the resulting range precision equated to approximately one foot of uncertainty in the elevations of the dataset, which was borne out when it was compared to another LiDAR dataset from a more precise sensor from a much lower altitude. A sensor that can only measure ranges in multiple decimeters of precision is not suitable for an engineering project, where centimeter level accuracy is desired.
Exploration of a number of publicly available datasets that matched projects acquired using a higher precision sensor revealed the same general results and caused a re-evaluation of how the high precision sensor could be employed. Shortcomings in high-altitude projects can be related to sensor ranging capabilities, survey control quality, geometry and density, and even the point density of the dataset. Lower density data is more difficult to control precisely, since it lacks horizontal definition of features, which are used to make adjustments to the data when compared to survey control points.
It is obvious that poor horizontal placement of the data has a substantial effect on the quality of the vertical accuracy. Translating this same issue to how conventional LiDAR datasets were controlled served as the impetus for further research.
Zoomed-in view of pixel detail from airborne LiDAR.
RESEARCH IN ACTION
Experimentation with high density, high precision datasetswas undertaken, employing dense, high quality survey control, and using a combination of design parameters, including software capabilities, optimized calibration boresight adjustments, high precision aircraft position and attitude tracking and input from system and software developers. Initial data acquisition was done using industry standard procedures that were modified slightly to test the various suggestions from the inputs described, with variations in flight line overlap, control density and placement, and direction of flight.
The first attempt was compared to a high density ground control network acquired using very labor and time intensive procedures conforming to existing specifications for supporting the design of runways. It is important to note, that the survey data required months of work, in highly hazardous conditions to acquire. The data acquisition for the airborne LiDAR system was accomplished in less than half an hour, mostly due to the complicated airspace constraints. There was not any undue risk done to the aircraft, or impact on the status of the active military runways.
The sensor employed for testing has a published range precision specification of <20-mm throughout its performance envelope and is capable of firing the laser up to 400,000/sec in a geometrically controllable pattern and recording up to five returns per pulse. When combined with operating altitude and speed of flight, the rate of acquisition of data is mostly easily related to two easily understood quantities: point spacing and point density.
The experimental dataset was flown at an altitude of 2,000-ft above the mean terrain of the site, at an average velocity of 112-kt—yielding a nominal point spacing of 1-ft on the ground and slightly more than seven points per square meter. The flight plan was designed to have areas of the runways covered with as few as one, and as many as three, strips of data collected. One runway was flown twice, in opposing directions; the other had only a single pass from end to end. Cross strips were flown to cover areas of interest with increased point densities.
Data processing consisted of all of the normal steps taken to convert, adjust and transform the point cloud data to the proper project horizontal and vertical datums. The smoothed best estimated trajectory (SBET) of the sensor platform was generated using a network of surrounding high data rate National Geodetic Survey Continuously Operation Reference Stations (CORS). The positioning system records the position of the aircraft five times per second, effectively determining its location and altitude every 37.6-ft of flight. Concurrently, the inertial measurement unit (IMU) records the roll, pitch and yaw of the aircraft two hundred times per second, with translates to every 0.9-ft of flight. These measurements are applied to the LiDAR and camera system data streams via the common time stamp provided by the GPS data to make adjustments for what the plane was doing when each laser measurement or photograph was acquired.
Individual strips of LiDAR data must be adjusted to match each other, using common data areas to compare trajectory and point positions to adjust for variations. Optimal comparative geometry is achieved by having directly opposing overlapping strips of data, followed closely by a single strip with cross strips. Once strip adjustments were optimized, they were transformed as a single entity to the site control points.
(Top) Intersection control and deviation plot with a table of min/max value variances. (Bottom) Overall control and deviation plot of project area.
TESTING THE RESULTS
Testing of the dataset, using a group of survey points acquired by an independent party, was undertaken to study the differences between the point cloud and approximately 18,000 ground survey points, which were acquired using the project specified procedures. Using the airborne data as a base surface due to its size and density, all of the ground survey points were compared to that surface and variations compiled. A plot of the variations, along with a spreadsheet tabulating the individual point differences from the base surface was statistically analyzed to ascertain mean values, standard deviation, minimum and maximum differences, and to group the results by magnitude.
Analysis showed that even in the initial test dataset, 70 percent of the ground survey points fell within 0.03-ft of the LiDAR surface, with the majority of the points exceeding the mean value of 0.0322 fell in areas lacking cross or redundant data, and less than optimal control geometry. All but a 14 points to be exact, fell within 0.10-ft of the base surface. Further analysis of the dataset revealed several areas where abrupt transitions in surface separation indicated there was some likelihood of small errors in the ground survey data.
The initial test served its purpose as a proof of concept, allowing the different flight patterns, control locations and overlap configurations to be validated, and leading to the acquisition of a second dataset.
This is comprised of a manufacturer recommended calibration area and data strips over two runways, each comprised of directly overlapping, opposing direction data. This is currently being controlled and survey data acquired for further testing, which is currently ongoing.