采用CP-GMM模型的实时人群异常检测

Real-time Anomaly Detection in Crowd Scenes Using CP-GMM Model

针对人群异常检测中出现的检测速度慢、检测精度低的问题,提出一种新的人群参数混合高斯模型(CP-GMM)算法来对人群参数进行建模分析,并通过筛选出小概率特征来检测人群监控视频中发生的异常现象。算法立足于人群中显著点,通过统计分析人群中显著点的特征来提取人群特征,并使用参数分析模型对感兴趣区域内速度大小、方向等参数进行异常检测。检测过程中不需要对人群中个体进行跟踪,也无需对检测模型进行大规模训练。实验表明,算法快速有效,在较低误报率下取得了较好的检测效果。

To improve the rate and performance of crowd anomaly detection algorithm,a novel Crowd Feature Parameter analyzing model,Crowd Parameter Gaussian Mixture Model（CP-GMM） is introduced.Based on this model,anomalies can be detected by checking out the crowd feature parameter with lower recurrence probability.The anomaly detection algorithm first gets good feature points in the crowd,then it counts and analyzes the features of the good feature points to extract the features of the crowd,and uses parameter analysis model to carry out anomaly detection on the parameters such as the magnitude and direction of the velocity.This anomaly detection algorithm needs not tracking individuals,and needs not large scale training of the detection model.Experiments demonstrate this anomaly detection algorithm is efficient,and can get better performance at lower false positive detecting rate.