Research On Ground Target Tracking Method Based On UAV Vision System

Target tracking technology based on the UAV vision system is widely used in aerial photography,search and rescue,reconnaissance and other fields because it barely rely on manual control and not restricted by signal transmission.In recent years,many robust tracking algorithms have been developed for classical visual tracking problems such as target scale change,local or global occlusion.However,due to the high complexity,most of them are not suitable for UAV real-time visual tracking tasks.To solve those problems,this research optimizes the Kernelized Correlation Filter(KCF)tracking algorithm to improve its robustness while reducing its time complexity;At the same time,this research builds an UAV processing platform based on RK3399 embedded development board,and verifies that the improved target tracking algorithm has good real-time robustness tracking effect through off-line testing and flight experiments.The main contents of this research are as follows:To solve the problem that KCF algorithm can not deal with the scale change of the target,the scale adaptive module is added and improved to the original algorithm.Firstly,the original HOG feature is replaced by FHOG feature to train KCF position filter,and the confidence of HSV color feature is added to the filter response to enhance the recognition ability of the algorithm.Secondly,the output of the position filter is used as the input of the scale filter to reduce the dimension of scale feature and the complexity of the algorithm.Finally,a large number of experiments are carried out to determine the minimum scale sampling number,to improve the speed of the algorithm on the basis of guaranteeing the ideal scale adaptation of the target.To solve the problem that KCF algorithm can not deal with the occlusion or loss of target,the target re-detection strategy is improved.The improved three frame difference method is used to filter the static background information in the image.Only the moving object regions are sent into the KCF filter for target re detection,to avoid the huge computation amount and improve the speed of the algorithm.Off-line test and actual flight test are carried out for the tracking algorithm.Firstly,the offline UAV123 dataset is used to test and verify the stability of the tracking algorithm,and the superiority of the tracking algorithm is proved by comparing with the typical target tracking algorithm.Finally,the actual flight experiment shows that the UAV tracking system in this research will hardly lose the target when the target scale changes and occlusion occurs while keeps the speed of 30 frames per second.