Comparing lidar sensor and platform performance for transportation

Establishing a Baseline for Technology

To determine the right mix of cost, resolution, and accuracy for project requirements, transportation planners need detailed knowledge of sensor and platform performance. In 2019, the ODOT and NV5 Geospatial set out to evaluate a variety of new sensors, deployed both on helicopters and UAVs, with the goals of:

• Understanding the horizontal and vertical accuracy of point clouds generated from three types of lidar sensors flown at different altitudes.

• Developing statistics for lidar accuracy on hard surfaces, such as asphalt and concrete, as well as soft surfaces, including bare earth and varying ground cover.

• Evaluating the qualitative aspects of the point clouds in terms of applicability for various project types, including fine feature determination of signs, rail, above ground utilities, lane markings and guardrails.
   

To achieve these goals, three applications were tested:

1. Traditional linear-mode RIEGL 480i sensor mounted to the belly of a helicopter.

2. NV5 Geospatial's comprehensive low-altitude sensor solution (CLASS), which combines two RIEGL VUX-LRs with nadir and oblique imagery. These also were mounted to a helicopter.

3. Single RIEGL VUX-1 flown on the small unmanned aerial systems (sUAS).

Flight parameter details were as follows:

For the different flights, the research team kept certain factors constant. They used the same centrally located high-accuracy global navigation satellite system (GNSS) base station for all trajectory post processing; the same software for creating point clouds with a lineup of all RIEGL sensors; and the same high-accuracy ground targets for calibration of all point clouds.

Success of the tests was also dependent on developing high-accuracy 3-D calibration and blind checkpoints on the ground throughout the two-mile project area. The ODOT surveying division used a mix of GNSS observations and differential leveling techniques to establish a large number of 3-D locations that enabled extensive testing in order to get a detailed understanding of the performance of the various platforms and sensors. The techniques included:

• 13 calibration points

• 15 hard surface (asphalt) blind QA points

• 36 cross section points on hard surfaces

• 344 cross section points off pavement in varying land cover

• 24 additional blind QA points on and off hard surfaces

Impressive Results

To evaluate accuracy of the three methods, the research team interpolated elevations at each of the surveyed XY locations from the three point clouds and compared them to the field elevations. These elevations also were compared to the elevations derived from the other two lidar point clouds, enabling the team to evaluate accuracy in six significant ways mentioned above.

The results for vertical accuracy, shown in detail in the chart below, were impressive:

• Hard surfaces, such as pavement and roadways: Accuracy was better than expected, and comparable across all platforms and sensors. The root mean square error (RMSE) was four hundredths of a foot, providing a 95% confidence interval of eight hundredths of a foot.

• Off-pavement areas, including bare earth, dense prairie grass and mesquite trees: Met expectations, but lidar points were consistently higher than ground surveys, resulting in reduced accuracy compared to pavement.

• Point clouds - In comparison, they were extremely consistent with one another with no anomalies among the tested areas.

Similarly the evaluation of horizontal accuracy exceeded expectations. Although the sample size was limited, each of the calibration targets was evaluated in the lidar intensity images and compared with their known horizontal position. This analysis provided a circular error at a 95% confidence level ranging from one- to two-tenths of a foot within the three independent point clouds.

Similarly, in overlaying the lidar intensity images over the digital orthophotos, produced at a ground sample distance (GSD) of 0.25 ft, there was perfect alignment at the pixel level, and no statistically significant difference between any of the point clouds.

Proving Value

As one of the first large-scale evaluations of these various cutting-edge lidar technologies and aerial platforms, this project has shown that both horizontal and vertical accuracy for airborne systems flown at altitudes of 250-600 ft are very accurate. More importantly, they offer results that are similar to those achieved by mobile or terrestrial collection on the ground.

In looking at the results, it is clear that helicopters may be ideal for large-scale projects covering long stretches of roadways or bridges. But this approach is not easily employed on a regular basis throughout the lifespan of the project.

Alternatively, transportation officials, project engineers and construction contractors could employ drones with lidar for smaller projects, such as the two-mile stretch analyzed by ODOT. It is much easier, quicker and more cost-effective to deploy drones for mapping of limited geography or when widening existing roads, since extreme accuracy and density can be achieved with the drones equipped with today’s impressive, small-format lidar sensors. Drones also could be used for continued monitoring of projects to ensure that plans are progressing as intended or adjustments are made in a timely manner.

The proven accuracy of using both helicopters and drones, with the right combination of sensors for the job, open up a plethora of options for transportation planning, engineering and construction. By gaining a better understanding of the advantages of each platform—helicopter or drone—and the sensor technologies that work best with them, transportation officials and contractors will have greater flexibility when it comes to designing and monitoring construction of new roadways, overpasses, and bridges.

About The Author: King is the Chief of Surveys for the Oklahoma DOT. Meade is a senior vice president at NV5 Geospatial. He is licensed as a professional engineer and professional land surveyor in multiple states.