嵌入式下公共场所暴力行为视觉检测方法的研究与应用

根据暴力行为发生时往往存在肢体冲突,并伴随着身体全身或局部部位出现较剧烈运动的特点,本文提出了一种基于运动图像的公共场所暴力行为自动识别方法,并能够移植到嵌入式设备中以便于分布式智能监控。首先,采用Lucas-Kanade(LK)光流法获得相邻两帧监控图像间的光流场,通过光流分析法确定各个运动目标的运动状态,并提取它们的光流特征值、光流速度和方向；其次,对各个运动目标的光流特征值、光流速度与方向进行统计分析,以掌握各个运动目标的运动趋势；最后,结合光流场变化情况及运动目标聚集状态判断是否存在暴力行为。实验结果表明,该嵌入式系统识别准确性好、实时性强、可靠性高。

Research and application of embedded visual detection methods about violence in public places

According that when the violence happen, there is a physical conflict and often accompanied by systemic or local-body strenuous exercise, we propose an automatic identification method for violence in public places. The method is based on the motion image technology, and can be transplanted to embedded devices to facilitate distributed intelligent monitoring. Firstly, use Lucas-Kanade (LK) to obtain the optical flow field of two adjacent frame images, specify the state of each moving object by analyzing optical flow, and extract their optical flow characteristic values, optical flow velocity and direction. Next, count and analyze the optical flow characteristic values, optical flow velocity and direction of each moving object, in order to grasp the movement trend of each moving object. Finally, according to the changes in the optical flow field and the aggregated state of moving objects to judge if there is violence. The experimental results show that the embedded system is good accuracy, real-time and high reliability for recognition.