Research And Implementation Of UAV Target Tracking System

With the continuous exploration of UAV technology by domestic and foreign researchers,the UAV control technology theory is widely used in practical scenarios and continuously improved.UAV are easily affected by environmental factors such as strong magnetism and gust of wind during flight,and target detection and tracking will be affected by changes in light and target shape,resulting in tracking failures.Providing good real-time performance and robustness play a key role in visual tracking.The research of this subject focuses on developing a set of UAV flight controller platform.Based on the self-developed flight controller platform,the key technologies such as data fusion,yaw compensation,target detection and target tracking for target tracking of UAV are studied.The main research contents are as follows:（1）Self-developed flight controller platform,including the layout design of the hardware,the underlying driver of the 10-axis core sensor and the code writing of the driver of the flight control platform peripherals,the use of gradient descent method for attitude calculation,and the use of complementary filtering to integrate gyroscope compensation attitude data.ADRC is used to control the flight of the UAV,and the real-time altitude data is calculated by integrating the acceleration,which enhance the stability and anti-interference performance of the UAV in flight,and lay the foundation for the stable tracking of the UAV with vision.（2）The tracking task consists of two modules,namely the detection module and the tracking module.The detection module uses the YOLO algorithm to identify the target and give the target frame.Aiming at the problem of missed detection and false detection when the YOLO algorithm detects small targets,the small target expansion method is used to enhance the target detection model.The tracking module adopts the Siam RPN algorithm introduced into the RPN network.Compared with the traditional multi-scale regression methods,the introduction of the RPN network in this algorithm improves the tracking speed,and divides the tracking process of the target into two branches: classification and regression,which improves the accuracy of the target.Box prediction accuracy and tracking speed.（3）Aiming at the problem of deviation in the yaw data solution caused by the influence of the magnetometer when the UAV is flying in a complex environment,it is proposed to improve the UAV yaw angle solution based on the data collection method of dual GPS antennas,solve and compensate in real time.The yaw angle of the UAV provides stable heading data for the UAV during the tracking process.（4）The target detection and target tracking nodes are designed based on the ROS system,and the target detection module,the tracking module and the UAV control module are connected through ROS communication to complete the task of tracking the target body.