Target Tracking Method Based on Kalman and Temporal Scale Adaptive Fusion CF

Abstract

Aiming at the problem of tracking drift and even loss caused by background occlusion in the tracking process of dim-small targets, a new tracking method combining Kalman and temporal scale adaptive fusion KCF (AKF-CF) is proposed.Firstly, extract the multi-modal features of the target grayscale image and directional gradient histogram (HOG), detect the target position, and improve the recognition ability of the target. Then the bilinear interpolation method is used to obtain a temporal scale adaptive box, and the size of the tracking box is dynamically adjusted. Using Average Peak Correlation Energy (APCE) for target occlusion detection and improving on KCF algorithm: if occlusion occurs, the Kalman filter algorithm is used to predict the target position at the next moment and replace the target in the original KCF algorithm to prevent tracking drift; If no occlusion occurs, the KCF algorithm continues to track, thus achieving effective tracking of targets in occluded environments. Using the OTB100 dataset for tracking long duration and occlusion sequences in videos. Compared with the ACSRCF algorithm on the OTB50 dataset, the accuracy and success rates improved by 0.8% and 0.06%, respectively. The data on the OTB100 dataset were 92.50% and 68.10%, respectively, surpassing other related filtering algorithms. Therefore, the success rate and tracking accuracy of the algorithm proposed in this paper have been significantly improved.