Visible Light Low-small-slow-target Tracking Algorithm Based on Improved MEANSHIFT

Due to the strong mobility and easy distortion of low-small-slow-target， the tracking and positioning error of these targets is large and the accuracy is low. Aiming at this problem， this paper proposes an improved meanshift algorithm for low small and slow target tracking algorithm based on multi feature fusion and region growth. Firstly， the ROI （region of interest） region is intercepted according to the manually calibrated initial position of the target， the gray histogram and hog （gradient direction histogram） features of the ROI region are extracted to establish a two-dimensional description template of the target. Then， constructe a new algorithm convergence criterion by combining the Bhattacharyya similarity coefficient between the target template and the candidate template and the Euclidean distance of Hu moment between ROI region and the candidate region. Finally， establish the template update mechanism by using the regional growth method to analyze the change of target area. Through the experimental performance with similar algorithms on the public data set LaSOT and four image sequences collected by ourselves， it shows that the algorithm in this paper is insensitive to the strong maneuvering and distortion low-small-slow-target， and the tracking effect is stable. Under certain constraints， the tracking accuracy of the algorithm can reach more than 95%， which has strong application value.