多特征博弈的目标跟踪算法

A Game Theory-Based Multi-Feature Fusion for Object Tracking

针对视觉目标跟踪领域中,采用单一特征的跟踪算法鲁棒性较差的问题,提出一种基于博弈论思想的多特征融合目标跟踪算法。在Mean Shift视觉跟踪框架下,将目标的颜色特征和运动特征作为两个博弈者,通过寻求二者博弈的纳什均衡,使不同特征对跟踪结果的贡献达到最佳平衡,进而更好地体现特征融合的优势。实验结果表明,该算法对目标剧烈运动、遮挡和背景多运动物干扰有较强的鲁棒性。通过基于博弈论的多特征融合方式在传统Mean Shift算法的基础上提出新算法,算法具有较好的跟踪性能。

Aimed on the problem that the robustness of single feature tracking algorithm is poor in visual target tracking, this paper proposes an object tracking algorithm based on multi-feature fusion using game theory. Under condition of the mean shift tracking framework, the paper takes color feature and motion feature as two players. Through looking for the Nash equilibrium of their game, the paper makes the contribution of different features in the tracking result the best balance, furthermore a higher advantage of multi-feature fusion. The experimental results show that this algorithm has the stronger robustness of tracking under object strenuous motion, occlusion and background motion interference. By means of multi- feature fusion based on game theory, the paper presents a new algorithm on the basis of the traditional mean shift algorithm, and the algorithm is good in performance on tracking.