Perception and Avoidance of Multiple Small Fast Moving Objects for Quadrotors With Only Low-Cost RGBD Camera

The autonomous navigation of unmanned aerial vehicles in a rapidly changing environment, such as avoiding small fast moving objects with onboard sensing, still remains a challenge. In this letter, we propose a complete system that only relies on a lightweight RGBD camera to achieve fast and accurate perception and avoidance of small dynamic obstacles, whereas navigating in a complex environment. Firstly, we detect the moving objects by Yolo-Fastest in RGB frame, obtain the 3D information with the depth image, and track the multiple detected objects with our proposed 3D-SORT (Simple Online and Real-time Tracking in Three-dimensional Space) algorithm. To achieve fast dynamic avoidance, we design an effective method to generate the optimized smooth trajectory to dodge all the static and dynamic obstacles with the predicted moving objects' trajectories. Finally, we integrate the above methods on our UAV platform, and demonstrate the performance of our system by testing thoroughly in simulation and real-world experiments.