视频监控中跌倒行为识别

Abnormal behavior recognition of fall in surveillance video

监控视频中的异常行为检测是计算机视觉研究领域的一个重要研究课题。人体跌倒行为作为异常行为的一种,可以对老龄化社会中的老年人跌倒行为做出实时预警,对保护老年人生命安全起到重要作用。本文采用三帧差法与更新运动历史图像相结合的方法获取运动前景,然后采用膨胀形态学操作与中值滤波操作,消除前景图像的噪声,对运动区域标记采用矩形包围框来获取感兴趣区域的形态变化,最后采用矩形框的宽高比、人体Hu矩特征、人体轮廓离心率、人体轴线角多特征融合来识别跌倒异常行为,对识别出的异常行为实时报警。实验结果表明对固定背景的监控视频中的单人跌倒异常行为识别,文中的算法具有很强的鲁棒性与稳定性。

Abnormal behavior detection in surveillance video is an important research topic in the field of computer vision. Fall belongs to one kind of abnormal behavior, which will play an important role in protecting the elderly if we can provide realtime warning for the fall behavior of the elderly. In this paper, three frame differencing method and the updating motion history method is combined to get the foreground. Then, dilation and median filtering are adopted to eliminate the noise of the foreground images. Next, the rectangular box is used to mark morphological changes to get the area of interest. Finally, the ratio of rectangular box＇s height and width ,the Hu feature,Contour eccentricity and Human axis angle are applied to detect the fall. Experimental results demonstrate the effectiveness and robustness of the proposed approach for the fixed background cases of surveillance video to detect the fall behavior of single person.