Air-to-Ground Active Object Tracking via Reinforcement Learning

Xin Liu, Weiya Ren, Jie Tan, Xiaochuan Zhang, Xiaoguang Ren, Huadong Dai

Abstract: Over the years, active object tracking has emerged as a prominent topic in object tracking. However, most of these methods are unsuitable for tracking ground objects in high-altitude environments. Therefore, the paper proposes an air-to-ground active object tracking method based on reinforcement learning for high-altitude environments, which consists of a state recognition model and a reinforcement learning module. The state recognition model leverages the correlation between observed states and image quality (as measured by object recognition probability) as prior knowledge to guide the training of reinforcement learning. Then, the reinforcement learning module can actively control the PTZ camera to achieve stable tracking and successfully recover tracking after object loss. Additionally, the study introduces a UE-free simulator that increases the efficiency of the training process by over nine times. High-altitude experimental results with the proposed method show significantly enhanced stability and robustness compared to the PID method. Furthermore, the results also indicate that the proposed method can significantly improve the image quality of the observation.