一种基于速度强度熵VMME与纹理特征的人群异常检测算法

人群异常事件检测是智能视频监控领域的重要研究内容, 文章提出了一种融合速度强度熵VMME与纹理特征的人群异常行为检测算法. 该算法采用LBPCM算法提取图像纹理特征, 在视频帧计算光流基础上, 获得特征点速度强度图, 并以其熵VMME作为场景运动特征, 将场景纹理特征和运动特征送入支持向量机训练分类. 实验表明, 新算法可实现对人群异常行为的检测, 且有较高准确率.

An anomaly detection algorithm based on VMME and texture features

In the field of intelligent video surveillance, the detection of abnormal events has remained an important subject in the research field. This paper proposes an algorithm for detecting abnormal behavior of crowds based on entropy of velocity magnitude map (VMME) and texture feature. Through the computation of optical flow in a video frame, the velocity magnitude map of feature points can be obtained, and the entropy of velocity magnitude map can be calculated as the feature of scene motion. LBPCM is first used to extract the texture features of the crowd, then the features of two kinds are fused into the support vector machine for training classification. Experiments show that the algorithm can effectively detect abnormal behavior and has high detection accuracy rate.