利用姿势估计实现人体异常行为识别

异常行为识别是近年来计算机视觉领域的研究热点。为了实现对多人体异常行为精确识别的目标,提出了一种基于人体姿势估计的异常行为识别算法。首先采用基于滤波通道特征的行人检测算法对各个目标人体进行定位;然后对每个人体构建基于图结构框架的外观模型;最终采用霍夫方向计算器算法(HOC)提取人体部件特征,从而进行行为分类。实验结果表明,该文算法可以在单帧图像上对多个人体的行为进行识别,并提供了多类别的异常行为分类,实验效果明显,准确率较高。

Abnormal human body behavior recognition using pose estimation

Abnormal behavior recognition becomes the recent research focus in the field of computer vision. In order to achieve the goal of recognizing multi human body abnormal behavior accurately, an abnormal behavior recognition algorithm based on human body pose estimation is proposed in this paper. Firstly, the pedestrian detection algorithm based on filter channel feature is used to locate each target human body, and then the appearance model of each human body is constructed based on pictorial structure framework. Finally, HOC (Hough Orientation Calculator) algorithm is used to extract the features of human body parts and classify the behavior. Experiment results show that the proposed algorithm can recognize multi human body behavior in a single frame image, and provide the classification of multiple kinds of abnormal behavior. In addition, the proposed algorithm obtains obvious experiment effect and high accuracy.