运动目标跟踪系统的遮挡问题处理  

Occlusion handling in moving object tracking system

作　　者：郑可飚[1] 黄文清[1] 张佐理[1] 李艳芳[1] 

机构地区：[1]浙江理工大学,浙江杭州310018

出　　处：《计算机工程与设计》2009年第11期2816-2818,共3页Computer Engineering and Design

摘　　要：遮挡是运动目标跟踪研究中的一个重要问题,介绍了基于Markov Chain Monte Carlo(MCMC)的运动物体遮挡问题解决方法。该方法通过建立贝叶斯模型,确定先验概率和条件概率,将车辆分割问题看成求后验概率最大时的车辆状态;然后运用MCMC方法对后验概率进行估计,设计MCMC标准对后验概率进行采样,用长方形模型来近似车辆外形。实验证明MCMC方法在不需对车辆单独初始化的前提下能有效的将相互遮挡的车辆分割出来,检测出车辆之间的相互遮挡。Occlusion is a significant problem for the research of moving object tracking. A MCMC-based method to solve the problem of occlusion is introduced. At first building Bayesian model, determine the prior probability and likelihood term. Formulate the vehicle segmentation problem as searching for the vehicle state, which maximizes the posterior probability. Then MCMC is applied to posterior probability estimation. A standard MCMC is designed to sample the posterior probability, and rectangle model is used to approximation the vehicle. Tests prove the MCMC-based method can segment multiple merged vehicles into individual effective. The method can automatically detect the vehicle inter-occlusion, without requiring a special region to initialize each vehicle.

关 键 词：运动目标跟踪 MCMC 分割 遮挡 贝叶斯模型 MEANSHIFT 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]