Multiplatform Stereoscopic 3D Terrain Mapping for UAV Localization

As part of an REU, it was given that the Department of Defense is interested in research on UAV homing in a GPS denied environment. This paper presents a study on a system of drones, in this case two, which are able to locate their absolute position when GPS becomes degraded or non-existent. The method to accomplish this starts when both robots capture the same ground image from different points of vantage. Since the distance between the UAVs (the baseline) is known and their altitude is known, computer vision is then used to generate a depth map. The results show that in general, an initial overlap of 78% was too little to generate disparity using the Semi-Global Block Matching algorithm, while an initial overlap of 92% was very usable. For analysis, 8 sets of 5 scenes were used to generate disparities. Using a Normalized Cross-Correlation Matching algorithm, the depth map was found on the global depth map and the location and orientation of the UAVs are calculated in every instance that a successful disparity map was generated.