基于三维虚拟警戒空间的异常入侵行为自动识别  

3D virtual warning region based abnormal invasion behavior recognition

作　　者：魏康[1] 管业鹏[1,2] 

机构地区：[1]上海大学通信与信息工程学院,上海200444 [2]新型显示技术及应用集成教育部重点实验室,上海200072

出　　处：《光电子．激光》2015年第9期1761-1767,共7页Journal of Optoelectronics·Laser

基　　金：国家自然科学基金(11176016;60872117);高等学校博士学科点专项科研基金(20123108110014)资助项目

摘　　要：针对目前视频异常入侵行为识别的不足,提出了基于三维虚拟警戒空间的异常入侵行为自动识别方法。基于人头检测与跟踪方法,根据视频监控场景中单一行人目标信息,建立行人三维平面方程,构建视频监控场景三维立体虚拟警戒空间,从而将行人是否进入二维场景警戒区域,转化为行人是否闯入三维立体虚拟空间,并基于行人头部投影射线的滑动滤波统计,实现行人是否入侵敏感保护区域的有效识别。所提方法不受设定警戒区域的规则形状限制,也无需对场景内容事先学习。对不同视频场景的实验验证及同类方法的定量对比结果表明,所提方法有效、可行。A novel method is developed to automatically recognize invasion abnormality based on a 3D vir- tual space,aiming at some limits in abnormal invasion behavior recognition for pedestrian in a video sur- veillance scenario at present. A 3D plane equation is constructed according to a single pedestrian extrac- ted and tracked by his head in a video surveillance scenario. A corresponding 3D virtual warning space in the surveillance scene is built. The problem whether a pedestrian invades warning region in a 2D scene is transformed to the one whether he intrudes the 3D virtual warning space. A sliding filter statistics strat- egy in ray projection of the pedestrian＇s head is developed to identify whether the pedestrian invades the warning protective region or not. Neither any regular shape constraint for the specified warning region or scene content previous learning is considered in the proposed approach. Some state-of-the-arts and exper- iments are done in some video scenes with different contents to test the performance of the proposed method in the same conditions. Experimental results show that the developed method is efficient and val- id without any specific hardware support or conditional constraint for the scenario.

关 键 词：异常入侵行为识别 三维虚拟警戒空间 压缩动态跟踪链 视频监控 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]