基于空间关联特征分布的目标跟踪

通过利用特征全局分布信息,提出一种以网格为数据单元的Mean Shift的目标跟踪算法(grid mean shift,GRIMSHIFT).针对传统Mean Shift目标跟踪算法的不足,GRIMSHIFT算法在m*n个像素的网格小区域内提取如颜色、角点量等局部区域特征值.在此基础上结合整幅图像进行约束Delaunay三角剖分得到图像像素间全局空间关联信息.在网格级上把局部特征信息和全局分布信息加权混合,使特征分布数据集具有了更高的目标辨识度；在视频序列中对动态网格特征分布连续运用Mean Shift便实现了对目标的跟踪.实验结果表明GRIMSHIFT拥有良好的实时性和准确性.

Tracking based on space feature distribution

After analyzing the information of the global feature distribution,a novel Mean Shift tracking algorithm using grid data units(grid mean shift(GRIMSHIFT)) is proposed.Aiming at the problems of the traditional Mean Shift tracking algorithm,GRIMSHIFT algorithm extract local area eigenvalue,such as color and corner energy,in the grid which content m*n pixels.On this basis,Integration the global information which reflects pixels spatial relationships extracted from constrained Delaunay triangulation of the image.Weighted integration the local characteristics and the global construction distribution of the image in the grid cell,so that the data set of feature distribution achieve a higher recognition.In the video sequence,using continuous Mean Shift on the dynamic feature grid distribution will achieve target tracking.The experimental results show that GRIMSHIFT has good real-time and accuracy.