视频中群体行为自动分析方法

Automatic Crowed Behaviors Analysis Method for Videos

针对监控录像中人群动静人体同时存在,人体之间的相互遮挡、聚集,导致人体检测和行为分析困难的问题,提出了一种基于头肩检测的自带反馈系统的人群群体行为分析方法。基于多视角头肩模型的头肩检测与基于Vibe的背景剪除方法相结合,检测结果一方面用来定位人体的位置和实现多目标跟踪,另一方面作为反馈因子来自动调节Vibe模型在不同区域的更新率,从而实现动静人体的同时检测,进而实现对人群聚集、流动的现象的自动分析。实验结果表明,该方法能有效分析人群的群体行为,并且快速鲁棒,能实现对监控录像的实时处理。

Due to the co-existing of moving people and stationary people,occlusions and gathering in crowed scenes,it is difficult to detect bodies of human and analysis crowed behaviors. To solve these problems,an analysis method for crowed behaviors based on head-shoulder detection and feedback loop is proposed. Human detection based on multi-view head-shoulder model and background removing based on Vibe are combined together.The result of detection is used for not only detecting human’s bodies,but also a feedback factor to adjust the update rate in different regions in frames,that would help to detect both moving bodies and stationary bodies,then implement an automatic analysis for crowed gathering and flow. Experimental results show that this method can analysis crowd behaviors effectively,and it is fast,robust and can process surveillance videos in real-time.