Research On Moving Target Tracking Algorithm Based On UAV Platform

At present,the target tracking technology is developed rapidly,and its application on the UAV platform shows great practical value,which has a profound impact on improving the intelligence and autonomy of UAVs.Especially in the military,target tracking technology can assist drones to complete a series of military activities such as enemy situation detection and target strikes,and has become a key technology to improve the performance of drones.There are many types of target tracking technology,but various problems encountered by the UAV during the tracking process,such as target scale change,target occlusion,and motion blur,cannot be completely solved.According to the characteristics of UAV platform tracking,the kernel correlation filter tracking algorithm is used as the framework of the tracking algorithm in this paper.The tracking algorithm is optimized from multiple aspects,and a set of target tracking algorithms with excellent performance and adaptable to the UAV platform are researched and designed.First of all,aiming at the complex and changeable environment of UAV tracking environment,an adaptive fusion strategy of three features of directional gradient histogram,color feature and local texture feature is proposed.By merging these three features,the effect of complementing target features in the detection process is achieved,and the ability of the tracking algorithm to deal with motion blur,light changes,and small deformations is improved.For the target occlusion problem encountered when the UAV is tracking,a strategy of "occlusion detection + offset step+ adaptive update" is proposed.The problems of partial occlusion or complete occlusion in a short time are solved from three aspects: judging whether the target is occluded,the processing scheme after occlusion,and implementing template adaptive update according to the degree of occlusion.Using the three-level scale pool alternating detection technology,the problem of constant scale changes in the UAV tracking process is solved.In the scheme,the limitation of the fixed aspect ratio of the target is broken,and the tracking algorithm has good tracking real-time performance.Secondly,when the UAV is tracking,it is necessary to manually select the tracking target in the first frame of image.More background information may be carried,and the entire tracking process is affected when selecting a target.Saliency detection technology is used to filter out background information.In current saliencydetection algorithms,color features are used as the basis for identifying salient regions.On this basis,spatial features and constraints are added to improve the accuracy of the saliency detection algorithm.At the same time,based on the saliency map obtained by the saliency detection algorithm,a scheme for extracting salient regions is proposed.The saliency algorithm and saliency area extraction scheme are used to achieve the goal of adapting the target area in the initial frame image.Finally,in order to verify the performance of the tracking algorithm,the OTB database and the actual aerial image sequence were used to analyze and test the tracking algorithm in quantitative and qualitative aspects respectively.Experiments show that not only the obvious superiority is achieved in qualitative experiments,but in quantitative terms,the tracking performance of other tracking algorithms is exceeded with 81.1% accuracy and 71.3% success rate.At the same time,the processing speed of the algorithm is 33 frames/s,and the real-time tracking requirements of the UAV can be met.