Learning Spatial–Temporal Background-Aware Based Tracking

Discriminative correlation filter (DCF) based tracking algorithms have obtained prominent speed and accuracy strengths, which attracted extensive attention research. However, some unavoidable deficiencies still exist. For example, the circulant shifted sampling process is likely to cause repeated periodic assumptions boundary effects, degrades tracker’s discriminative performance, target not easy locate in complex appearance changes. In this paper, a spatial–temporal regularization module on BACF (background-aware filter) framework proposed, performed by introducing temporal deal effectively with effects issue. At same time, of recognition improved. This model can be optimized employing alternating direction multiplier (ADMM) method, each sub-problem has corresponding closed solution. addition, terms feature representation, we combine traditional hand-crafted features deep convolution linearly enhance performance filter. Considerable experiments multiple well-known benchmarks show proposed algorithm performs favorably against many state-of-the-art trackers achieves an AUC score 64.4% OTB-100.