Target tracking method based on the fusion of structured SVM and KCF algorithm

Abstract

Aiming at the drift problem of traditional KCF (kernelized correlation filter) target tracking algorithm caused by scale change and fast movement, a tracking method based on the fusion of structured SVM (support vector machine) and KCF algorithm is proposed. Combined with the location and scale information of the tracking target, a structured SVM classification and prediction model is established to replace the ridge regression filtering method in KCF, train the samples collected by cyclic matrix, and solve the problem of not detailed classification. In addition, a search strategy optimization method based on tracking target motion features is proposed to reduce unnecessary search time in dense sampling, further improve the search efficiency and classifier accuracy of traditional KCF algorithm in dense sampling, improve its calculation efficiency and target tracking performance in complex environment. This can solve the problem that the target cannot be accurately tracked due to the drift caused by the change of target scale and fast movement. Experiments show that the algorithm can locate the target area more accurately when the target scale changes and fast movement. Compared with the traditional KCF tracking algorithm, it has better performance and stronger robustness.