Search Our Resource Collection
Here you can find articles about our products and services, case studies that we have conducted, announcements and updates, white papers and recorded webinars that explain and demonstrate various tasks and workflows that can be performed in our software.
All | Articles | Case Studies | News | Recorded Webinars | White Papers
Ground Classification in True View EVO
Author: Lewis Graham, February 17, 2020
Collection of Surface Topography
One of the more popular uses for the True View 410 3D Imaging System (3DIS) is in the collection of surface topography for site planning. The data from the UAV LIDAR/Imagery sensor replaces data that might typically be collected using conventional topographic surveying techniques such as measuring ground elevation using a Global Navigation Satellite System (GNSS) Real Time Kinematic (RTK) rover (so-called “pogo” collection).
The primary workflow for topographic modeling (once accuracy has been assessed and deemed adequate) is separation of ground from everything else and then the generation of a gridded elevation model (and/or topographic contours).
True View 410: Drone Mapping Software Included
One of the huge advantages of any True View 3DIS is the inclusion of our True View Evo, post-processing software (based on our ubiquitous LP360 point cloud software). Not only does True View Evo do all the True View 3DIS post-processing (from raw data to colorized point cloud), it also includes a rich set of tools for analyzing point cloud data and generating a set of output products.
Topographic Workflow
The general workflow for doing a topographic project is:
- Plan the mission such that you have good overlap between flight lines with the LIDAR data limited to a ±40° field of view (FOV). If the terrain is hilly, use 3D mission planning so that a relatively constant height above ground is maintained. Until we add mission planning to True View Evo (slated for late 2020), we recommend Litchi.
- Set out a few check points so you can assess vertical accuracy and, if necessary, debias the point cloud (“Z” bump)
- Fly the mission.
- Post-process the data to a geocoded point cloud (colorized if you have a 3DiS)
- Assess accuracy
- Classify low (“noise”) points so they will not be included in the ground model
- Automatically classify ground
- Inspect and clean as necessary
- Thin data using statistical thinning tools (e.g. a median filter)
- Export grid and/or contours
As an example, we will run ground classification on a highway project we recently flew for the Alabama Department of Transportation (AL DOT). During a heavy rain, a section of Alabama route 231 was heavily damaged by ground sliding under the pavement. We were asked to do a True View 410 flight and provide a ground model.
Examining the cross section of Figure 1, note the high vertical relief of this project. The grid shown in the figure represents 25 feet intervals. This project had over 500 feet of vertical relief in the relatively small mapping area. Clearly this requires 3D mission planning.
Ground Classification with True View 410
As an example, we will run ground classification on a highway project we recently flew for the Alabama Department of Transportation (AL DOT). During a heavy rain, a section of Alabama route 231 was heavily damaged by ground sliding under the pavement. We were asked to do a True View 410 flight and provide a ground model.
Examining the cross section of Figure 1, note the high vertical relief of this project. The grid shown in the figure represents 25 feet intervals. This project had over 500 feet of vertical relief in the relatively small mapping area. Clearly, this requires 3D mission planning.
Figure 1 – High vertical relief project
The ground classification algorithm used in our drone mapping software, True View Evo, is an “Adaptive Triangulated Irregular Network (TIN)” classifier. The general algorithm goes as (it is fully automated, so you do not manually carry out the below steps):
- Create a “virtual” grid over the data of some specified size (we used 50 feet for this project)
- Find the lowest point in each grid and change its class to Ground
- Construct a TIN of these sparse Ground points
- Loop through all points that have not been classified as ground and test against the TIN surface. If the point is not too far from the surface and does not make too steep an angle with respect to the surface, add it to the Ground class
- Repeat step 4 until no need points are being added to the Ground class or you are satisfied with the density of the ground class
The adaptive TIN ground classifier is extremely popular due to its ability to adapt to many types of terrain without a lot of user tweaking. However, its Achilles heal is the first step – classify the lowest points in each grid cell to ground. If this lowest point happens to be low noise rather than ground, the following steps of the algorithm will perform poorly in the region of the noise point.
We were very careful in selecting the Quanergy M8 Ultra as the scanner for the True View 410 (and the Riegl mini-VUX2 for the soon to be released True View 615/620). This scanner exhibits exceptionally low noise, aiding in this ground classification scheme. That said, True View Evo does have an automatic noise classification algorithm that can be run on the data to mark any low points.
Once low points have been addressed, the automatic ground classifier is executed against the data. In general, the results are remarkably good with not a lot of data cleanup required. Figure 2 depicts the result of an automated run of the True View Evo ground classifier on the project data. Note that these data are being displayed in a really nice hybrid view mode enabled within True View Evo: all points not classified as ground are shown in RGB color and ground is displayed in orange.
Figure 2 – Ground Classified Data
Finally, in Figure 3 is depicted 10 feet contours produced in True View Evo superimposed over a Google satellite imagery backdrop. You can see by the steepness of the terrain what a difficult and long project this would be if ground survey techniques were employed.
Figure 3- 10 ft. contour superimposed on a Google backdrop
All in One Drone Mapping Solution
This sort of scenario shows the value of a True View 410 as a rapid response system. In less than 8 hours total time (including mobilization, data collection and post-processing) we have a ground model of current conditions on the ground. The really nice thing about this is the software is all one single package: no need to cobble together workflows from a disparate collection of applications.
Type: Articles
Categories: Resources
Topics: 3Dis, Accuracy, Data Processing, ground classification, LIDAR, sUAS mapping, True View, UAV LIDAR
