Discriminative Scale Space Tracking Research Articles

Retrogression of correlation filters for discriminative visual object tracking

Although discriminative correlation filter-based (DCF) trackers have shown excellent performance in recent years, it still has the problem of model drift. A generic approach alleviating model drift is proposed for DCF tracking. Specific correlation filters are learned in feature channels and the combination of response maps depends on the confidence level of corresponding channels. Model drift is eliminated by allowing contaminated correlation filters to retrogress to the ones that are not influenced by impure samples. As a result, correlation filters are purified by neglecting the impure part of them. Meanwhile, both the confidence level of detection and the variant level of interframe feature space are considered during online update of correlation filters. Our approach is experimented on both OTB-50 and OTB-100 datasets. Especially on OTB-100, our approach outperforms the baseline fast discriminative scale space tracking by 6.9%, 8.1%, and 5.6% in mean distance precision, mean overlap precision, and the area-under-the-curve score, respectively. Moreover, the competitive performance is presented by comparing with other state-of-the-art tracking algorithms.

Discriminative Scale Space Tracking.

Accurate scale estimation of a target is a challenging research problem in visual object tracking. Most state-of-the-art methods employ an exhaustive scale search to estimate the target size. The exhaustive search strategy is computationally expensive and struggles when encountered with large scale variations. This paper investigates the problem of accurate and robust scale estimation in a tracking-by-detection framework. We propose a novel scale adaptive tracking approach by learning separate discriminative correlation filters for translation and scale estimation. The explicit scale filter is learned online using the target appearance sampled at a set of different scales. Contrary to standard approaches, our method directly learns the appearance change induced by variations in the target scale. Additionally, we investigate strategies to reduce the computational cost of our approach. Extensive experiments are performed on the OTB and the VOT2014 datasets. Compared to the standard exhaustive scale search, our approach achieves a gain of 2.5 percent in average overlap precision on the OTB dataset. Additionally, our method is computationally efficient, operating at a 50 percent higher frame rate compared to the exhaustive scale search. Our method obtains the top rank in performance by outperforming 19 state-of-the-art trackers on OTB and 37 state-of-the-art trackers on VOT2014.