一种基于位移概率矩阵的目标跟踪方法  

A novel object tracking approach based on matrix of displacement probability

作　　者：蒋文斌[1] 邵志远[1] 叶启伟[1] 

机构地区：[1]华中科技大学计算机科学与技术学院,湖北武汉430074

出　　处：《华中科技大学学报（自然科学版）》2006年第z1期45-48,共4页Journal of Huazhong University of Science and Technology(Natural Science Edition)

摘　　要：针对非刚体目标跟踪问题,提出了一种采用快速块运动估计算法SSC-APDS计算位移概率矩阵的新方法.由SSC-APDS首先计算出可能位移的误差和(SAE)曲面,并据此形成位移概率矩阵.在此基础上,采用一个基于位移概率矩阵的时间滤波器来完成对目标特征块的运动分析.最后根据各特征块的运动情况,得到目标的整体运动.实验证明该跟踪方法不仅具有很好的鲁棒性,且计算效率高,在以视频网格为平台的视频监控系统中跟踪效果好.Focusing on the non-rigid object tracking problem,this paper presented a new approach for the computation of the matrix of displacement probability(MDP) based on fast block motion estimation algorithm subsampling search candidates in adjustable partial distortion search(SSC-APDS).The surface of the sum of abstract error(SAE) of possible displacements is worked out by it,which leads to MDP.Then an MDP-based time filter was employed to the tracking analysis of every featured block.Finally,the object tracking was obtained according to the integration of all tracking of blocks.Experiment showed that this approach not only has higher robusticity,but also can reduces computation in the object tracking in video surveillance based on video grid.

关 键 词：目标跟踪 位移概率矩阵 块运动估计 时间滤波 视频网格 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]