3D Imaging is Different
Published On: July 08, 2020
Our True View LIDAR/Imagery fusion systems for drones are designed to create point cloud data that have been accurately colorized with Red-Green-Blue (RGB) camera data. We call these "3D Imaging Systems (or Sensors)", 3DIS. We have now been building and using these systems for over a year. We have two basic members in the True View 3DIS family; the True View 410, a Quanergy laser scanner based version for general mapping and the True View 615/620 which use the RIEGL miniVUX2 laser scanner, aimed at survey-grade work.
We have recently engaged in a project with the US Department of Agriculture to explore the value of various types of data for inspecting earthen dams. As part of this work, we have been collecting some exemplar data using our 3DIS as well as other sensors on a few USDA managed dams. Something that we have started taking for granted, but USDA found remarkable, was that we can fly a site in the morning and post a colorized point cloud that same afternoon. An example of a view of a True View 410 3DIS data set is shown in Figure 1. This depicts a bit of the infrastructure area of a dam site. The data are an exquisite 3D colorized point cloud that has a metric accuracy better than 5 cm RMSE with no ground control.
Better Data and Faster Processing!
One of the biggest advantages of a True View 3DISâ„¢ is the speed with which these data can be produced; post-processing typically takes about 2/3 the flight time. Thus a 15 minute data collect can be processed to a colorized point cloud in about 10 minutes. Trying to do this same process using imagery to point cloud processing (so-called structure from motion, SfM) would not only take hours of post-processing but also does not have the vegetation penetration of LIDAR. Thus the True View 3DIS workflow has that very rare combination of better data and faster processing!
Figure 1 :View of True View 410 3DIS data set
Choosing the Right Platform
We do a lot of data collection flights in support of our internal research and development, qualifying the sensors we build for resale and in work with select customers, developing best practices (our USDA work being a prime example). We used to have a step in planning where we would select the best sensor for a particular job - should we use an Inspire 2 with an X4S camera, a Sony DSLR, a LIDAR unit sans camera (back when we resold YellowScan systems) and so forth. Now, the True View 410 is the universal solution to most projects. If we cannot use LIDAR because the ground was too wet (or too dark such as coal piles) then no worries; we just process the data from the True View 410 dual cameras through an SfM program such as Pix4D or Agisoft. Of course, now with the introduction of the True View 615/620, we will need to think about which class of 3DIS we need to deploy - general purpose or survey grade?
True View EVO: Quick Workflow
One final note on rapid processing. Each True View sensor includes a copy of True View EVO, our post-processing software. EVO provides a plethora of tools for processing colorized (and non-colorized, for that matter) point cloud data. For example, Figure 2 shows a series of cross-sections created along the berm of the dam. The view is from under the berm since this really showcases the fit of the automatic cross-sections to the point cloud data. This workflow was all performed within True View EVO. The general steps were:
- Process to a colorized point cloud
- Classify low noise (there is seldom low noise with the True View 410)
- Classify ground using EVO's automatic ground classifier
- Draw a center line where the profiles are to be centered
- Run the automatic cross-section tool to create profile lines along the center line, draped to the point cloud surface (filtered to ground for this example)
The above workflow takes less that 30 minutes on this exemplar data set, even with a bit of manual ground cleanup thrown in!
Figure 2: Cross-sections of the berm of the dam
Processing Data: Time is Money!
The True View EVO workflow for post-processing saves literally hours of time as compared to SfM processing. A more subtle savings is the colorization attributes of the point cloud. Classification of a non-color point cloud can be quite difficult - is it a pile of dirt or is it a brush pile? One belongs in the ground class whereas the other does not. It's very tough to tell without color. The takeaway here is the old adage "time is money." True View 3DIS respect the value of your time!