摘要
很多传统视觉监控的研究工作集中于行人跟踪、行为和事件检测、步态或人脸识别等,然而角色识别却研究较少.针对多摄像机监控中角色识别的应用问题,该文作者提出了一种基于贝叶斯因果网的角色识别方法.该方法不仅用到了通常的一些人物视觉特征,而且还考虑了时间特征、空间统计特征和一些其它特征.作者将这些特征向量的概率分布参数化,特征向量成员之间的因果关系通过有向无环图的方式来表达,然后通过提取的特征来计算概率以识别人物角色.实验的结果证明了方法的有效性.
Many works of conventional surveillance are focused on people tracking,behavior or event detection,gait or face based recognition,etc.However,role identification is also very important but usually paid less attention.In video surveillance,video analysis and video data mining,it is better for us to treat detected or tracked people with different strategies considering their roles.This paper proposes a multi-camera system to identify people with specific roles using a causal network.Not only visual features but also spatio-temporal features,and some object specific features are involved in the new system.Multiple cameras benefit locating the position of moving objects and overcoming occlusions.Experimental results demonstrate the effectiveness of the method.
出处
《计算机学报》
EI
CSCD
北大核心
2010年第12期2378-2386,共9页
Chinese Journal of Computers
基金
国家"八六三"高技术研究发展计划项目基金(2009AA01Z305)
国家自然科学基金(60833009
60903072)资助~~
关键词
监控
多摄像头
角色识别
贝叶斯因果网
surveillance
multiple cameras
role identification
Bayes causal network
