Research on Unmanned Aircraft Environment Perception Methods in Urban Low Altitude Complex Backgrounds

With the continuous development of drone-related technologies, the application of UAV (Unmanned Aerial Vehicle) in urban low-altitude scenes is gradually increasing. Autonomous localization and perception of the surrounding environment are prerequisites for autonomous control. As a result, our team apply the VDO-SLAM algorithm to the scene of coexistence of dynamic and static obstacles in the complex background of urban low altitude. Effective segmentation of static background and moving objects is achieved by combining semantic segmentation and optical flow estimation algorithms. Since there is a lack of relevant public data for urban low-altitude scenarios, a binocular image dataset containing both static and dynamic objects is created by Airsim simulator, which is utilized for training a semantic segmentation model. In addition, a novel numbering maintenance algorithm is introduced for the semantic segmentation of continuous image sequences. The experimental results show that the algorithm can realize the autonomous perception of UAV posture and the motion state of dynamic objects in low altitude complex urban environment. This method provides the necessary self-pose information and reliable environmental data for autonomous obstacle avoidance of UAVs, and provides important support for the application of UAVs in urban low-altitude environment.