Study On The Detection And Tracking Of Ground Targets By Using UAV

Vision sensor is one of the main sensors of UAV,and Visual target detection and tracking play a key role in the tasks of UAV navigation,military reconnaissance,field rescue and traffic monitoring.The visual target detection and tracking on the UAV platform is different from the traditional monitoring system.The flexible maneuverability of the UAV will cause the image sequence obtained by the airborne sensor device to be complex and changeable,and the target scale,angle,and other changes will be dramatic.Frequent occlusion and other issues make the detection of visual target detection more difficult,decreasing the detection rate,and reducing tracking stability and continuity.What’s more,the large image size and obvious noise will reduce the real-time performance of the algorithm.In view of the above problems,the research work of this paper is based on aerial video of UAV,focusing on the issues of visual target detection,single-target and multi-target tracking.Based on this,using Qt to develop interactive interface software,functional testing of the algorithm and visual representation of algorithm results in real time.The main works done are as follows:1)Based on the characteristics of UAVs to obtain image sequences and aiming at the problem that static background modeling cannot obtain effective target detection,this paper proposes the detection of moving targets based on optical flow.the FAST feature points are detected,the optical flow is used to complete the tracking,and the perspective transformation is used for image registration.The feature points are also used for cluster analysis,and using multi-frame difference to obtain the approximate target area,then using image processing methods such as connected domain and morphological filtering to further improve the detection results.2)The current tracking algorithms based on correlation filtering and other types of tracking algorithms with better performance are studied,which the theoretical analysis and actual tracking performance of the algorithm are also analyzed.In order to deal with the problem of mutation and occlusion caused by the UAV’s perspective,the color information is integrated into the HOG feature to improve the tracker’s stability and anti-blocking performance.3)The current multi-target tracking has encountered data association and target detection problems.This paper adopts a multi-target tracking algorithm based on tracking segment which is slightly different from the mainstream,uses LBP feature and Adaboost learning algorithm to train the target detection model,and combined with single-target tracking algorithm,using the classifier to complete the initialization,and using the single target tracker to track the target one by one to,achieving multi-target tracking for UAVs.4)Researching interactive interface development,using C++ programming language and Qt development framework to develop software interactive interface,integrating various target detection,single-target tracking,multi-target tracking and background modeling algorithms into the software framework,and providing algorithm testing for each function displaying the tracking and detection effects of the algorithm in real time.