一种基于视觉注意机制的改进粒子滤波跟踪算法

摘要

目标跟踪是计算机视觉研究中的一个重要课题,它是目标行为理解的基础,是图像系统连续准确工作的重要部分.针对单一特征跟踪算法识别准确性不高,特别是在遮挡状况下无法对目标特征进行检测和跟踪的问题,考虑到粒子滤波算法在处理非线性、非高斯跟踪问题上的优越性,提出一种融合颜色、纹理和运动信息等多类特征的改进粒子滤波跟踪算法；并参考人眼的视觉注意机制,根据目标物体在不同场景下对人眼刺激的显著性不同,对目标的各个特征按照显著性强弱排序,并以此对散布粒子进行过滤.与单一特征和多特征目标跟踪算法的对比实验表明,所介绍的算法比基于单一特征的目标跟踪算法具有更高的准确性和鲁棒性,且比多特征跟踪算法的实时性更好.%Object tracking is a significant subject in the research of computer vision; it is not only the foundation for target behavior understandings; but also an important contributor for an image system to work continuously and accurately. Aiming at the inaccuracy at recognition for single feature tracking algorithm, especially when shaded, it is impossible to detect or track object features. Considering the superiority of particle filter algorithm at non-liner and non-Gaussian tracking problem handling, the paper proposes an improved particle filtration tacking algorithm that merges multiple features such as color, texture, and motion etc. ; meanwhile, referring to human eyes visual attention mechanism, according to different significances to human eye stimulation by an object in different scenarios, sorts the object' s features by significance strength so as to filter distributed particles. Comparative experiments between object tracking algorithms with single feature and multiple features indicate that the proposed algorithm is better at accuracy and robustness than single feature tracking algorithm and better at timeliness than multiple features tracking algorithm.