基于ICLBP的多特征自适应加权融合目标跟踪算法  

A Target Tracking Algorithm of Adaptive Multiple Feature Weighted Fusion Based on ICLBP

作　　者：李小强 李亚飞 LI Xiaoqiang;LI Yafei(Henan Nc Application Technology Engineering Research Center,Zhengzhou 450046,China;School of Information Science and Engineering,Henan University of Technology,Zhengzhou 450001,China)

机构地区：[1]河南省数控应用技术工程研究中心,河南郑州450046 [2]河南工业大学信息科学与工程学院,河南郑州450001

出　　处：《信阳师范学院学报（自然科学版）》2021年第3期478-483,共6页Journal of Xinyang Normal University(Natural Science Edition)

基　　金：国家自然科学基金项目(31171775);河南省教育厅高等学校重点科研项目(17A510008)。

摘　　要：针对Mean Shift跟踪算法中使用单一的特征对目标进行描述而导致跟踪算法鲁棒性不高的问题,提出了一种多特征融合的目标跟踪算法.该算法选取HSV颜色特征和ICLBP纹理特征,建立目标模型的概率密度.根据目标区域确定背景区域,计算不同特征对目标和背景的区分性度量值,并以此设定和更新特征融合权值.使用特征融合权值系数建立多特征描述的目标模型,在Mean Shift算法框架上实现目标跟踪.结果显示,该算法对背景干扰和部分遮挡具有较好的鲁棒性.与传统的Mean Shift跟踪算法相比,跟踪效果有所提高,鲁棒性更好.A target tracking algorithm of multiple features fusion is proposed for the problem that the tracking algorithm of single feature describing targets in Mean Shift tracking algorithm is not robust enough.The proposed method uses the HSV color feature and the ICLBP texture feature to establish the probability density of the target model.The background area is determined according to the target area,and the feature fusion weights are set and updated based on the distinguishing measure of the target and the background under different features.The proposed algorithm uses the feature fusion weighted coefficient to establish a multi-feature description of the target model,and achieves the target tracking in the Mean Shift algorithm framework.The results show that the proposed algorithm has good robustness to background interference and partial occlusion.Compared with the traditional Mean Shift tracking algorithm,the tracking effect is improved and the robustness is better.

关 键 词：目标跟踪 Mean Shift 区分性度量 多特征融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]