摘要
针对自助银行服务中出现的暴力事件,提出了一种基于目标空间位置关系和事件语义模板的检测方法。该方法对划定区域选用背景自适应更新的背景减法得到侧影图,并统计人体个数;采用粒子滤波和均值漂移相结合的机制,跟踪人体并计算相关距离;将侧影图与异常事件的语义模板,采用快速模板匹配方法及改进伪Zernike矩构建的一个7维形状向量实现相似匹配来判定是否出现异常事件。实验结果表明,事件检测准确性较高,能够较好地满足实际应用需求。
This paper presents a new method based on spatial relationship among objects and event semantics template to detect violent event in self-service banking. The method gets the silhouette by the background subtraction algorithm in which the background updates adaptively, and counts the number of the people in the video. Then, it tracks the people using particle filter and mean shift method, and computes the distance between the persons. The fast event semantics match template method and a shape description vector consisting of seven improved pseudoZernike moments （IPZM） are combined to detect abnormal event. Experimental results show the method is effective for the practical application.
出处
《计算机科学与探索》
CSCD
2012年第4期377-384,共8页
Journal of Frontiers of Computer Science and Technology
基金
重庆市自然科学基金No.CSTC2007BB2454
重庆邮电大学博士启动基金No.A2009-25~~
