Obstacle detection algorithm for unmanned aerial vehicles using binocular stereoscopic vision

摘要

Unmanned Aerial Vehicles are currently gaining widespread adoption in multiple roles. In this paper, we proposes obstacle detection algorithm for unmanned aerial vehicles using stereoscopic vision. Depth of objects in the image could be calculated. There is also another important property of image captured by a stereo camera; if an object is at a certain position it will appear at a certain combination of location in images from left and right camera but not any other. This property can be used to detect obstacles by checking a number of sample point placed in a grid in front of the drone. For each point we then use the property stated above to check if locations in the image specified by the property matches. We have evaluated the effectiveness of this algorithm. This algorithm scored 17 out of 20 cases within acceptable time constraints, which demonstrated its worthiness for further developments.