Research On Ground Target Tracking Algorithm Based On Drone Platform

The rapid development of computer and communication technology has promoted the progress of unmanned aerial vehicle(UAV)technology,which is also used in more and more industries because of its flexibility and rapidness.Tracking of ground targets is an important UAV application,and is widely used in emergency search and rescue,border reconnaissance,security and so on.In order to solve the problem of the inability to automatically locate and track ground target,the ground target tracking algorithm based on the UAV platform is designed and studied in this paper,the ground target is automatically detected,tracked and located,and the UAV is guided to continuously track the target.The specific research work of this paper is as follows:1)The design of tracking algorithm is carried out in this paper,divided into detection,visual tracking,positioning guidance.Detection module is for automatic identification of the target in the view of the optoelectronic platform camera,if the target is detected then the result will be obtained as the input of visual tracking module to initialize tracking operations.By calculating pixel difference between the position of view center which is also position of the camera aiming line,and the center of the target,the camera will be controlled to aim target throughout the process.By combining drone’s GPS,gimbal corner information and the laser ranging module information,the relative position between the drone and the tracked target is calculated through the method of alignment coordinate conversion.And the target’s GPS is also calculated according to the relative position information.The relative position information will help drone for flight decisions during tracking missions.2)The UAV field of view is large,the target background is often very complex,and subject to the performance of the airborne terminal processor,the existing target detection algorithm is difficult to meet the dual requirements of detection accuracy and speed.This paper proposes an improved algorithm based on the YOLO framework.The calculation complexity of improved algorithm is greatly reduced under the premise of ensuring detection effect by replacing the original algorithm backbone network with lightweight and efficient MobileNet feature extraction network.And the description ability of target feature of the network is improved by using feature attention module SE-Block,which makes up for the loss of precision caused by replacing the feature extraction network.After testing algorithms on the VOC data set,the improved algorithm is faster than the original algorithm by 75.6%in detection speed,can achieve 10.6 FPS running speed on the UAV airborne TX2 processor,and the detection accuracy can meet the application needs of target tracking,and has a better balance between detection accuracy and speed.3)In order to improve the accuracy and anti-occlusion performance of visual tracking algorithm,an improved algorithm based on the SiamRPN is proposed.Firstly,using the previous frame target interference information to strengthen the feature similarity response function,the algorithm can make more full use of strong negative samples of the previous frame,improve the ability to distinguish between the tracked target and the strong interference.Secondly,Kalman filter is used to predict the trajectory of the target and take the prediction position of the current frame as the target search center,and further improve the target tracking accuracy and anti-occlusion performance.Tested on the UAV 123 dataset,the improved algorithm enhanced tracking accuracy by 7.3%,the tracking success rate increased by 10.2%.The improved algorithm has 13 FPS running speed on TX2 processors,and has better anti-occlusion performance.This paper first proves the improvement of the enhanced algorithm by testing on the public dataset,and finally proves that the algorithm can realize ground target tracking under the UAV platform.