Research on Target tracking Based on Improved KCF Algorithm

Abstract

Aiming at the problem of tracking instability caused by imperfect feature expression and variable scale, an improved KCF algorithm is proposed in this paper. Firstly, HOG and MB-LBP features are extracted to form H-L fusion feature, and Hue color features are extracted too. Then the output response values of extracted features are calculated respectively by position filter. Next, the fusion is carried out by linear weighting at the feature level, to obtain the target’s predicted location. And in the tracking process, the scale changes of targets are estimated and adaptively adjusted by the standard deviation of the feature matching corner points of the two frames before and after. Finally the difference of the adjacent two frames is judged by the corner matching condition, and the position filter is updated, to achieve better target tracking effect. The experiments on OTB-50 and OTB100 show that the algorithm not only improves the tracking accuracy by 8.3% and the success rate by 10.5%, achieves the real-time tracking effect.