综合颜色和小波纹理特征的免疫粒子滤波视觉跟踪  

An immune particle filter video tracking method based on color and wavelet texture

作　　者：韩华[1] 丁永生[1,2] 郝矿荣[1,2] 

机构地区：[1]东华大学信息科学与技术学院,上海201620 [2]数字化纺织服装技术教育部工程研究中心,上海201620

出　　处：《智能系统学报》2011年第4期289-294,共6页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金重点资助项目(61134009);国家自然科学基金资助项目(60975059;60775052);国家ITER计划国内配套研究资助项目(2010GB108004);教育部高等学校博士学科点专项科研基金资助项目(20090075110002);上海市优秀学术带头人计划资助项目(11XD1400100);上海市科学技术委员会重点基础研究资助项目(10JC1400200;09JC1400900);上海市科学技术委员会技术标准专项资助项目(10DZ0506500)

摘　　要：针对标准粒子滤波视觉跟踪时的缺陷,尤其是其"样贫"和目标特征不明显等问题,提出了结合小波纹理特征的免疫粒子滤波算法.小波纹理特征的加入,使得单纯依靠颜色特征不能很好适应环境变化的情况得到了改善.同时通过加入免疫优化算法,提高了粒子的多样性,尤其是在发生遮挡时减少了"样贫"的影响.通过实验验证了所提算法的有效性和鲁棒性.与基于单一特征的跟踪方法相比,该算法能稳健地实现对复杂场景下目标的跟踪.Focusing on the drawbacks of standard particle filters in visual tracking, a new immune particle fiher was proposed based on color and wavelet texture feature to avoid ＂sample impoverishment＂ and incomplete feature. Since the wavelet feature was added, the poor performance of only one feature was improved when the environment changed. In addition, the proposed immune optimization algorithm improved the diversity of particles, especially the decline of the impact of sample impoverishment when occlusion occurs. At last, the effectiveness and robustness of the proposed method was verified by experiments. In contrast to the single feature tracking method, the proposed method can track targets with complex backgrounds more steadily.

关 键 词：粒子滤波 免疫粒子滤波 小波纹理特征 相似性度量 视觉跟踪 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]