基于纹理特征的尺寸方向自适应均值漂移算法

2018-07-16

作者：吴晗，董明利，樊凡，郑浩单位：北京信息科技大学光电测试技术北京市重点实验室

摘要：针对均值漂移算法在目标跟踪中由于相似背景干扰和固定带宽值核函数引起的跟踪精度下降的问题，结合目标的纹理特征及尺寸方向的变化，提出了一种新的跟踪算法。首先，通过自适应算法求解出目标的灰度阈值，得到纹理特征参数，并结合颜色特征建立目标模型；然后，通过求解目标相似度最大值提取出运动目标的图像区域；最后，采用协方差矩阵估计方法，结合最小二乘算法对运动目标尺寸方向的变化量进行自适应计算，表示出运动目标在视频帧中的像素区域，保证了跟踪的准确性。实验证明，本文提出的算法较其它均值漂移算法有更高的精度和效率。

关键词：均值漂移；纹理特征；灰度阈值自适应；尺寸方向自适应

中图分类号：TG806；TH741；TP391.41文献标志码：A

Scale and Orientation Adaptive Mean Shift Tracking Based on Texture Feature

Wu Han，Dong Mingli，Fan Fan，Zheng Hao

Abstract:There are two problems about classical meanshift algorithm which are caused by background similarities of color feature and scale and orientation change of the target in the process of target tracking，a scale and orientation adaptive mean shift tracking based on texture feature is proposed on this paper.Firstly，this paper solves the threshold value of gray level by adaptive algorithm to get the model of target combining with color feature to establish target model.Secondly，this paper extracts the image area of target through the method of template matching.Finally，for ensuring tracking accuracy，the scale and orientation adaptive covariance matrix estimation and the least square algorithm are used to satisfy the realtime scale and orientation changes of the target.Compared with other classic mean shift algorithm by experiments，the algorithm in this paper has improved significantly in precision and efficiency.

Keywords:mean shift；texture feature；threshold of texture feature adaptive；scale and orientation adaptive