中文摘要：

针对视频场景的人群运动状态分析问题提出了一种方法,包括人群密度分级和运动异常检测。该方法利用场景中不同区域的亮度信息作为BP网络的输入向量分类人群密度,降低了计算的复杂性,排除不必要的干扰信息。在异常检测方面利用光流法获取人群的运动信息,包括运动速度和运动方向。实验结果表明,该方法的精度及实时性均高于传统方法,对确定视频场景中人群运动状态是有效的,可以为防止大规模安全事故提供参考。

英文摘要：

This paper deals with the analysis of pedestrian motion in video and proposes a method including pedestrian density classifying and motion exception detection. The method utilizes luminance information from different areas of the video scene as the input of BP neural network to classify the density of pedestrian which reduce the complexity of computation and excludes the unnecessary information. This paper also uses optical flow method counting pedestrian motion velocity and its direction. The method shows its advantage in both accuracy and effectiveness. It can be used as a reference to catastrophe prevention.