Abstract: How to effectively analyze and monitor sudden crowd abnormal condition has become a research hotspot issue at present. The most common way to detect the crowd abnormal motion is to analyze the change of the speed. However, in addition to the speed, apparently the direction of the motion is another important feature of the velocity. Studies show that when the crowd motion is abnormal, the most significant phenomenon is the speed or the direction of the crowd motion suddenly changes. For instance, from slow walking to fast running, sudden change of the motion direction. Based on this, this paper proposes a new algorithm of detecting the crowd abnormal motion that is based on acceleration. The algorithm uses the pyramid Lucas-Kanade optical flow method to analyze the crowd motion by calculating the “frame average acceleration”. The experimental results show that the proposed algorithm can rapidly and effectively detect the crowd abnormal motion. Moreover, compared with the method of only using the speed to detect, the results of the proposed algorithm are more accurate and the range of applications will be more extensive.

Key words: crowd abnormal motion, pyramid Lucas-Kanade optical flow, frame average acceleration

摘要： 如何有效地分析、监测突发人群异常状况已成为目前的一个研究热点问题。通过速度值的大小变化来分析是否出现异常状况是目前最常用的人群异常检测方法。但是，除了值的大小之外，运动的方向显然也是速度的另一个重要的特征。研究表明：人群出现异常时，一个最明显现象是人群的运动速度大小或运动方向突然出现变化，例如，由慢速行走转变为迅速奔跑、突然改变行进方向等异常运动。基于此，提出了基于加速度的人群异常检测算法。该算法采用金字塔Lucas-Kanade光流法，通过计算“帧平均加速度”对人群的运动进行分析。实验结果表明所提出的算法能够快速、有效地检测出人群运动的异常，而且相对于单纯运用运动速度值大小进行检测方法，检测结果更准确，应用范围更加广泛。

关键词: 人群异常运动, 金字塔Lucas-Kanade光流法, 帧平均加速度