A Simulation Framework for Vision-Based Target Tracking Control of UAVs

This work focuses on visual-based target tracking of drone via visual simulation system. A simulation platform is particularly designed to simulate the environments in real-world applications. The platform follows a modular design. In this platform, Unreal Engine is chosen as the rendering platform, Airsim is employed to construct sensor models and acquire the kinematic state of drone body and environmental information. The drone is controlled by PX4 autopilot, and trajectory is planned by EGO-Planner. The target recognition and tracking is achieved by a siamese-based tracking algorithm. Four flight simulation experiments are conducted to verify the use of the developed framework for vision-based tracking, which demonstrate the usefulness and applying method of the simulation system for UAVs in disturbance scenarios.