一种自动分割跟踪视频运动对象的方法  

A method for automatic segmentation and tracking of video moving object

作　　者：杨文佳[1] 窦丽华[1] 张娟[1] 

机构地区：[1]北京理工大学自动控制系模式识别与智能系统研究所,北京100081

出　　处：《光学技术》2009年第2期276-279,283,共5页Optical Technique

基　　金：省部级基金资助项目(40405020202)

摘　　要：针对视频图像中单个运动对象的分割和跟踪问题,提出了一种基于时空离散度的视频对象分割跟踪算法。该算法首先采用时空离散度进行运动对象区域的粗定位;然后通过形态学算子来进行细定位,并获取运动物体的二值模型;再然后使用部分Hausdorff距离实现后续帧中运动对象模型的匹配;最后为了容纳运动对象在视场中形状的变化,在图像距离空间中引入了欧几里德范数作为约束条件来完成运动物体模型的更新。实验表明,在运动对象跟踪过程中相邻位置差未超过2个像素点,从而验证了算法的有效性。To solve the problem in segmentation and tracking the moving object in video sequence, a novel moving object segmentation and tracking approach is proposed based on spatial-temporal dispersions. Firstly spatial-temporal dispersion is computed to detect the moving object region, and mathematical morphology is used to detect the edge of the moving object. Then the matching of the object is judged by partial Hausdorff distance, which can be computed with a distance transform model. Finally, a restraint condition based on Euclid norm is used to update the object model. Especially in the whole object tracking system, the tracking error does not present above 2 pixels. So the proposed approach is proved to be feasible and effective.

关 键 词：视频运动对象 时空离散度 形态学算子 HAUSDORFF距离 图像匹配 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]