基于最大似然的分布式多视角目标跟踪算法  

A Distributed Multi-view Object Tracking Algorithm Based on Maximum Likelihood Criterion

作　　者：冯巍[1] 杨成[1] 胡波[1] 冯辉[1] 

机构地区：[1]复旦大学电子工程系,上海200433

出　　处：《光电工程》2009年第10期16-23,共8页Opto-Electronic Engineering

基　　金：国家重点基础研究发展973计划项目(2006CB705700)

摘　　要：针对多摄像机视频跟踪的应用场景,提出了一种具有分布式计算特点的目标跟踪算法。该算法首先利用最大似然准则对多视角目标跟踪问题进行建模,并对似然函数进行分解,最后借助消息传递机制高效求解出跟踪目标在各个摄像机视野中位置的估计值,而不再需要将图像数据传输到中央处理单元进行集中式处理。同时,通过融合多个视角的图像信息,有效地抑制了遮挡现象给跟踪结果造成的影响。定性和定量实验均证明了该算法的有效性。A novel algorithm is proposed to perform object tracking with multiple cameras in a distributed computing manner. Firstly, the maximum likelihood criterion is employed to model the multi-view tracking problem. Then, the likelihood function is factorized based on the conditional independent assumptions. Finally, the estimation of the target location in each camera view is obtained by using an efficient message passing mechanism. In our algorithm, image data from all cameras no longer need to be transmitted to and processed by a central unit. Meanwhile, by taking full advantage of image data from multiple cameras, the tracking algorithm is very robust to occlusion. Both qualitative and quantitative experiments have demonstrated the effectiveness of the proposed algorithm.

关 键 词：分布式多视角目标跟踪 最大似然准则 消息传递机制 归一化互相关 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TN911.73[自动化与计算机技术—计算机科学与技术]