Robust Visual Tracking via Iterative Gradient Descent and Threshold Selection
Zhuang Qi, Junlin Zhang, Xin Qi
TL;DR
The paper tackles the challenge of precise target estimation in visual tracking under outliers by introducing IGDTS, a robust regression framework that combines a Gaussian–Laplacian noise model with an iterative gradient descent and threshold selection procedure. The core objective, $L(\beta,\gamma) = \tfrac{1}{2}\|y - \beta X - \gamma\|_2^2 + \sum_i \lambda_i |\gamma|_{(i)}$, uses a sorted-L1 penalty to selectively suppress outliers, and the associated IGDTS algorithm guarantees monotonic decrease toward a robust solution. IGDTS is extended to a generative tracker via the IGDTS-distance $d_{IGDTS}$, enabling principled state estimation under an affine motion model and an IPCA-based appearance model, with an update scheme to adapt to appearance changes. Extensive experiments on 14 challenging datasets demonstrate that IGDTS outperforms existing trackers in accuracy and robustness, offering a practical, efficient approach for robust visual tracking in real-world scenarios.
Abstract
Visual tracking fundamentally involves regressing the state of the target in each frame of a video. Despite significant progress, existing regression-based trackers still tend to experience failures and inaccuracies. To enhance the precision of target estimation, this paper proposes a tracking technique based on robust regression. Firstly, we introduce a novel robust linear regression estimator, which achieves favorable performance when the error vector follows i.i.d Gaussian-Laplacian distribution. Secondly, we design an iterative process to quickly solve the problem of outliers. In fact, the coefficients are obtained by Iterative Gradient Descent and Threshold Selection algorithm (IGDTS). In addition, we expend IGDTS to a generative tracker, and apply IGDTS-distance to measure the deviation between the sample and the model. Finally, we propose an update scheme to capture the appearance changes of the tracked object and ensure that the model is updated correctly. Experimental results on several challenging image sequences show that the proposed tracker outperformance existing trackers.