Binocular Vision-based Autonomous Path Planning for UAVs in Unknown Outdoor Scenes

It is a classic task for Unmanned Aerial Vehicles (UAVs) to accomplish autonomous scene perception and path planning in unknown 3-D outdoor environments. This paper investigates the problems of obstacle avoidance and path planning using binocular vision system. During the UAV's flight, the binocular vision sensor is used to obtain the local environment information in real time, and the distribution of obstacles in the environment can also be analyzed with the depth images acquired by the binocular vision. Inspired by the idea of dynamic window algorithm, a 3-D path planning algorithm is proposed to convert the global path to the combination of a group of local paths by using a series of 3-D models of predefined local paths. According to the screening algorithm for the passable candidate paths, the UAV will select the optimal one to guide its flight. A series of experiments are conducted by using a quadrotor platform DJI M100 and experimental results show the validity of the proposed approach.