Stereo vision-based fast obstacles avoidance without obstacles discrimination for indoor UAVs

A stereo vision-based obstacle awareness and avoidance algorithm for indoor UAVs is described in this paper. While in most papers UAVs perceive the environment by vision-based obstacle discrimination, in our method a scene depth map is directly used, which makes our algorithm more adaptive in complex environment with numerous obstacles. For the purpose of lower time cost and less match mistakes, edge information is used to improve original area-based stereo matching method. Furthermore, the perceived environment is represented by a grid-based depth map, according to which the optimal guide point is chosen. Finally, a feasible avoidance path is generated by adding way points while comparing the depth of hypothetic way points with the depth of corresponding grids. Experiments results show the effectiveness of our algorithm.