A Vision-based UAV Tracker Aiming at Aerial Targets

Autonomous unmanned aerial vehicle (UAV) tracking has been widely used in inspection, rescue, follow-up and many other fields. Compared to ground targets, aerial targets are smaller, more flexible and possess more freedom of motion, which makes the task more challenging. In this paper, we propose a vision-based UAV tracker aiming at aerial targets. Firstly, a lightweight detection model is deployed for target detection and target's position is estimated taking advantage of detection box information and camera internal parameters. Then, a flexible motion prediction method is proposed followed by an efficient and occlusion-aware front-end path searching approach. Finally, an optimization-based multicopter trajectory planning framework is employed to generate a spatio-temporal optimal trajectory. The experimental results indicate that our tracker achieves great robustness and security in both vertical and horizontal tracking.