Current methods of large scale terrain modeling can be cost and time prohibitive. We present a method for integrating low cost cameras and unmanned aerial vehicles for the purpose of 3D terrain mapping. Using structure from motion, aerial images taken of the landscape can be reconstructed into 3D models of the terrain. This process is well suited for use on unmanned aerial vehicles due to the light weight and low cost of equipment.
– Attitude Information via Photogrammetry
The attitude and position of the helicopter while taking images is represented by the blue rectangles. This demonstrates that the GPS/INS information is not needed for creating the map, but, rather, extraction of both the map and the attitude + position of the helicopter is possible simultaneously.
The following models were generated from flights at the White Sands Missile Range (WSMR) in Las Cruces, New Mexico. They were created in real-time, using vision based 3D mapping. Covering an area approximately 2km x 1km, the UAV flew at an altitude of 600m AGL. The point cloud models display the geometric locations of interpolated points. Models with images draped over are also seen below.
– Point Cloud Model
WSMR – Las Cruces, NM
– Model with Image draped over
Las Cruces, NM
– Point Cloud generated from Las Cruces flights
Model generated at White Sands Missile Range –
Comparisons of this technique have also been performed. The image below compares DEMs produced at ranging resolutions. All DEMs were generated using vision based, 3D mapping techniques. The area shown is 33.4592 N, 115.8565 W in the Salton Sea State Recreation Park in Palm Springs, CA.
– Aerial photo of region- 10 cm DEM
– 25 cm DEM
Vision based models from this area were also compared to LIDAR datasets. The image below shows a DEM generated from each technique.
