基于交叉熵稀疏表示的鲁棒视觉跟踪算法  

Robust visual tracking based on correntropy sparse representation

作　　者：丁维福[1,2] 张讲社[2] 

机构地区：[1]北方民族大学数学与信息科学学院,宁夏银川750021 [2]西安交通大学数学与统计学院,陕西西安710049

出　　处：《辽宁工程技术大学学报（自然科学版）》2017年第8期883-891,共9页Journal of Liaoning Technical University (Natural Science)

基　　金：国家973计划项目(2013CB329400);国家自然科学基金(61075006);宁夏自然科学基金(NZ12209)

摘　　要：为研究复杂视频环境下目标的有效跟踪问题,在粒子滤波框架下,提出了利用稀疏表示的方法学习有效外观模型的鲁棒视觉跟踪算法.与经典的稀疏跟踪器不同,该方法通过给跟踪目标中被遮挡的像素和奇异值分配较低权值,而给目标像素分配较高权值,有效地解决了跟踪过程遮挡、阴影和噪声问题.为了进一步提高跟踪器的性能,对目标模板集实现动态更新.使用EMD度量了模板集和候选目标的相似性,可进一步改善遮挡问题.将本文提出的算法在复杂的视频序列上与5中流行的跟踪器进行了比较,实验表明,本文提出的算法在性能、精度及鲁棒性方面都显示了优越性.To study the efficient methods in the complex visual circumstance, this paper presents a novel online object tracking algorithm with sparse representation for learning effective appearance models under the particle filtering framework. Compared with the state-of-the-art sparse tracker which simply assumes that the image pixels are corrupted by independent Gaussian noise, our proposed method, which is based on correntropy sparse representation, is much more insensitive to corruptions which assigns small weights to the occluded pixels and outliers. To further improve the tracking performance, target templates are dynamically updated to capture appearance changes. In our mechanism of template updating, the similarity between the templates and the target candidates is measured by earth movers＇ distance. Two ensemble methods constructed from five state-of-the-art trackers with the individual trackers were empirically compared. The proposed tracking algorithm runs in real-time and performs favorably against state-of-the-art algorithms on challenging sequences in terms of efficiency, accuracy and robustness.

关 键 词：鲁棒视觉跟踪 交叉熵稀疏表示 自适应外观模型 粒子滤波 遮挡和奇异值 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]