Research On Aerial Video Target Detection And Tracking Algorithm Based On DSST

Target detection and tracking has been a hot research topic in the field of computer vision in recent years,and various new algorithms and ideas have emerged one after another.The application of moving target detection and tracking based on aerial video is more and more important not only in the military,but also in the daily life of ordinary people,so it has high research value.At the same time,detecting and tracking moving targets in aerial video is also a very challenging task,because aerial video has problems such as low image resolution,simultaneous movement of background and target,small target size and occlusion.In this thesis,the target detection and target tracking algorithms under aerial video are studied separately to solve the difficult points encountered in the target detection and tracking process,and improve the accuracy and real-time of the algorithm.The main work of the thesis is as follows:Aiming at the problem that the background and the target move together in the aerial video,this thesis proposes an improved three-frame difference method based on background compensation to accurately extract the target.The SURF feature matching algorithm with good performance is selected to extract the feature points.Then the RANSAC algorithm is used to estimate the homography transformation model of the motion background for background compensation.Due to the existence of holes and noise in traditional frame difference method,this thesis proposes an image quality assessment algorithm to evaluate the quality of multiple frame difference results and add it into the three-frame difference method to obtain more accurate moving targets.Moreover,the edge detection algorithm and morphological processing are integrated together to further boost the overall detecting performance.The extensive empirical evaluations on aerial videos demonstrate that the proposed detector is very promising for the various challenging scenarios.In view of the limited computing power of the UAV airborne equipment used for the shooting target,considering the robustness and real-time performance of the target tracking,this thesis selects the DSST target tracking algorithm as the base algorithm.The scale search strategy of DSST tracking algorithm is improved,and the D.S.C dimension update strategy is proposed.The efficiency of the algorithm is improved from the aspects of reducing the data dimension and reducing the search step.At the same time,for the aerial video,there are many interference items,and the target is easy to occlude when moving.The multi-peak target re-detection strategy is designed to prevent the model drift caused by similar targets or background noise,and improve the positioning accuracy.Aiming at the problem that most algorithms can't track moving targets for a long time,this thesis proposes a long-term moving target tracking algorithm.Based on the DSST target tracking algorithm,the target confidence calculation strategy,model update strategy and target re-detection mechanism are designed to ensure the reliability of algorithm tracking during long-term target motion.The experimental results on the DARPA VIVID aerial video data set prove that the proposed DSST-based long-term target tracking algorithm is superior to existing tracking algorithms such as KCF,SMAF,CSK and TLD,both in accuracy and robustness.There are significant advantages on the top.The algorithm is as high as 75.5% and 93.3% in distance accuracy(DP)and overlap precision(OP),and its center position error(CLE)is only 10.9.The algorithm tracks moving objects in aerial video in real time at a rate of about 30 frames per second.