Research on UAV Multi-Obstacle Detection Algorithm based on Stereo Vision

The UAV has the advantages of small size, low cost and flexibility, and has a wide application prospect in military, civilian and scientific research. While applying the drone to perform the mission, it is of great significance to ensure that the drone detects obstacles during the flight. Due to the high maneuverability of UAVs, detection algorithms based on single-target often miss other potential obstacles, resulting in insufficient reliability. This paper proposes multi-obstacle detection algorithm based on stereo vision. The dual-camera platform is established and the image computational hardware is applied to run the algorithm. After 3D reconstruction and stereo matching, the depth map of the scene is obtained. After that, a series of image processing algorithms are applied to find out five subtle dangerous objects and give them bounding boxes. At last, this paper establishes the hardware system and carries out several tests outsides. According to the multi-obstacle detection algorithm proposed in this paper, the results show that the algorithm can detect at most five obstacles in 15m and has practical application value.