基于LHMM的聚众异常事件实时检测改进方法

An improved method based on LHMM for gathering abnormal event realtime detection

提出了一种改进的分层隐马尔科夫模型(LHMM)结合熵值的聚众异常事件实时检测方法。使用直方图均衡化对视频帧做预处理,增加图像质量;以分块区域中的人数和总速度作为观察值分两层训练出聚众事件的LHMM。当观察值序列与模型的相似度大于设定阈值时,利用光流法计算该帧熵值,当熵值大于设定阈值时,则认为聚众事件发生;否则,为非聚众事件,继续下一帧的处理。大量实验结果表明:改进的方法具有较高的识别率、较好的鲁棒性和高的处理速度,并且应用环境更广。

An improved layered hidden Markov model （LHMM）and entropy value of real-time detection method for gathering abnormal event detection is presented. The video frame is preprocessed using histogram equalization to increase image quality ; the number of people and the total speed of people in block area as the observed values, then gathering event＇ s LHMM in two levels is trained. When the similarity between observation sequences and the model is greater than certain threshold, then calculate the entropy of the frame by using optical flow method. When the entropy value is greater than the certain threshold, a gathering event is judged to happen;otherwise, it is not a gathering event and continues to process next frame. A large number of experimental results show that the proposed method has higher recognition rate, better robustness and high processing speed and wider applications.