Robust DCF object tracking with adaptive spatial and temporal regularization based on target appearance variation

Abstract

To alleviate the boundary effect and constrain the update of the tracking model, current object tracking methods based on Discriminative Correlation Filter (DCF) usually introduce spatial and temporal regularization constraints in the filter training objective function. However, these regularization constraints with fixed coefficients greatly limit the adaptability of the tracker with respect to target appearance variation. This paper proposes a spatial-temporal regularization model based on the real-time target appearance variation for the filter training, improving the adaptability of the filter related to target appearance variation. Moreover, the filter training objective function with the adaptive spatial-temporal regularization is proposed to enhance the robustness of the filter. Finally, an iterative optimization method based on the alternating direction method of multipliers (ADMM) is proposed to update the filter, and the convergence proof of the optimization method is also presented. Comparison experiments with some representative trackers including ASRCF,ARCF, CSR_CF, DSAR_CF and SSR_CF etc. on OTB2015, UAV123 and LaSOT databases show that the proposed algorithm effectively improves the tracking accuracy.