摘要
群体骚乱行为突发性强、破坏性大,是视频监控关注的重点。研究群体骚乱行为的智能检测方法有助于提高视频监控的智能化水平。现有行为检测方法在检测群体骚乱行为时虚警率较高,实用性较差。因此,依据群体骚乱行为发生时光流模值变化大的特性,提出了一种基于光流模值变化频率的群体骚乱行为检测方法。该方法首先计算视频中每一帧图像上各像素点的光流;然后自适应求取反映像素点光流变化大小的二值映射图;接着分区块计算视频片段上的光流变化频率直方图,构建行为描述子;最后采用线性支持向量机进行特征训练与分类。实验表明,所提方法在检测群体骚乱行为时虚警率和漏警率低、识别率高,可广泛用于智能视频监控领域。
Group riot activity is the focus of video surveillance,which is emergent and destructive.To detect the group riot activity intelligently is helpful for improving the intelligence level of video surveillance.The common activity detection methods have high false alarm rate while detecting group riot activity from videos.In this paper,a detection method for group riot activity was proposed based on change frequency of optical flow＇s magnitude.Firstly,the optical flow of each pixel on each frame in a video is calculated.Secondly,a binary map is obtained to reflect the changes of optical flow aptively.Then an activity descriptor by dividing an image into several blobs and the change frequency histogram of optical flow is calculated independently.Finally,the activity features are trained and classified by using a linear support vector machine.Experiments show that the new method has low false alarm rate,low missed alarm rate and high genuine acceptance rate,while detecting group riot activity.So it can be widely used in the field of intelligent video surveillance.
出处
《计算机科学》
CSCD
北大核心
2016年第5期283-287,共5页
Computer Science
基金
广东省科技厅项目(2013B070207001)资助
