隐式马尔科夫HDP非参数贝叶斯视频异常检测

Hidden Markov Model Based Non Parametric Bayesian Algorithm For Video Anomaly Detection

为提高视频异常检测算法的适用性,同时提高算法的识别效率和精度,提出一种基于隐式马尔科夫非参数贝叶斯算法的视频异常检测算法。首先,针对跨越多天的视频画面进行片段序列划分,然后对所有帧的像素位置进行光流矢量总数统计和特征向量提取;其次,针对流数据分割和隐藏模式发现过程中模式数量未知问题,利用分层Dirichlet过程的隐式马尔科夫算法和非参数贝叶斯因子分析进行视频数据流分割和模式发现;最后,引入了一个交互式系统,允许用户检查和浏览可疑事件。实验结果显示,所提算法可实现视频异常的自适应模式发现,并可提高算法的识别精度和效率。

In order to improve the applicability of video anomaly detection algorithm,and improve the recognition accuracy and efficiency of the algorithm,we proposed an hidden markov model based non parametric bayesian algorithm for video anomaly detection.Firstly,the video sequence is divided into several segments,and then the total number of optical flow vectors and the feature vectors are extracted for all the pixels of all frames;Then,in order to solve the problem of mode number unknown in current data segmentation and hidden patterns,we used the Hidden Markov algorithm of the hierarchical Dirichlet process and nonparametric bias factor analysis to make video data stream segmentation and pattern discovery;Finally,an interactive system was introduced to allow users to check and view suspicious events.The experimental results show that the proposed algorithm can realize the adaptive pattern discovery of video anomalies and improve the recognition accuracy and efficiency of the algorithm.