基于运动轨迹的监控场景分析模型  

Surveillance Scene Analysis Model Based on Motion Trajectory

作　　者：梁浩哲[1] 李国辉[1] 张军[1] 

机构地区：[1]国防科技大学信息系统工程重点实验室,长沙410073

出　　处：《计算机科学》2011年第9期264-266,303,共4页Computer Science

基　　金：国家自然科学基金项目(60902093)资助

摘　　要：所提出的模型通过层次分类聚类过程学习运动轨迹信息结构。用拓扑先验分类轨迹的相似性运动空间,然后结合混合模型拟合各运动特征的统计分布,得到潜在运动规则;最后基于规则模型检测监控场景中的异常运动行为。轨迹多维运动线索的统计模型对噪声具有较强的鲁棒性,同时利用先验分类使运动规则具有较明显的语义结构。实验结果验证了模型的有效性。The proposed model learns the potential structure of motion trajectory within a hierachical process composing of classification and clustering.Topology prior was used to classify the spatial similarity of trajectory.After that by utilizing Mixture Model to estimate distribution of motion features,the potential motion rules of the scene were obtained,based on which abnomal motion can be detected.The model is robust to low-level noises because of the statistic learning of multi-dimentional motion clues,and by combining prior the motion rules have obvious semantic structures.The real surveillance experiment results verified the efficiency of the proposed model.

关 键 词：视觉监控 混合分布估计 场景理解 异常检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]