Autonomous Landing Point Retrieval Algorithm for UAVs Based on 3D Environment Perception

Abstract

With the development of science and technology, UAV has been applied to all aspects of life. Computer vision technology has been used as a common external sensing technology because of its low cost, high reliability and good accuracy. The autonomous landing of UAV is an important technology of autonomous flight. The common visual guidance landing of UAV is based on the ground target detection of cooperative target and the output of UAV landing point position. This method requires the UAV to accurately identify the established ground cooperative target, so when the target is lost or the UAV needs to make a forced landing, the UAV will not be able to land normally, in serious cases, it will lead to UAV damage. Therefore, in order to solve the problem of UAV autonomous landing in unknown environment, this paper proposes an autonomous landing point retrieval algorithm based on UAV 3D environment perception. First, this paper will conduct a preliminary road search at a certain height to obtain information on both sides of the road. Secondly, the external 3D point cloud environment is reconstructed in real time through the binocular camera, the pre-landing plane is found by plane fitting, and the 3D information is mapped to the 2D plane to extract the plane mask, and finally the random forest is used to determine the landing point. Through experimental analysis, the algorithm in this paper can better guide the UAV to land in an unknown environment.