Object Tracking Based on a Time-Varying Spatio-Temporal Regularized Correlation Filter With Aberrance Repression

When used for object tracking, the discriminative correlation filter (DCF) is effective, but its performance often burdened by undesirable boundary effects. Meanwhile, when there too much background information in training samples of DCF, it will be easier to learn area deviating from tracking object. Further, illumination variation, partial/full occlusion, and appearance variations, render response map aberrance (CF) more prone occur. To overcome these problems, an model based on a time-varying Spatio-temporal regularized with repression proposed this paper. Firstly, adding term traditional CFs limit change rate generated detection phase, tracker can obviously repress maps; secondly, adjusting regions suitable high confidence scores spatial reliability map, effectively overcomes adverse effects caused effect; finally, introducing temporal term, also has superior ability partial occluded objects those large variations. Significant experiments OTB100, VOT2016, TC128, UAV 123 datasets have revealed that thereof outperformed many state-of-the-art trackers DCF deep-based frameworks terms accuracy, success rate, A-R rank, etc.