Robust Autonomous Unmanned Aerial Vehicle System for Efficient Tracking of Moving Objects

The autonomous unmanned aerial vehicle (AUAV) system has gained increasing attention due to its wide range of applications in various fields. However, integrating complex deep learning perception algorithms with autonomous navigation system on limited computing resources of UAV poses a significant challenge. In this paper, we propose a solution by leveraging the co-design of software and hardware in a heterogeneous computing system on the UAV, maximizing hardware resources including sensors. Building upon an existing autonomous navigation system, we integrate and optimize deep learning perception algorithms to create an autonomous, robust, and stable AUAV system for tracking moving objects. By achieving autonomy control with the limited resources on the UAV, we extend the usability, offering new possibilities for various domains such as agriculture, search and rescue, and infrastructure inspection.