摘要
为了提高实时性和精确度,提出一种利用角点动能检测群体异常行为的方法.首先,利用金字塔Lucas-Kanade光流法计算FAST(Features from Accelerated Segment Test)角点光流,筛选出运动的角点;然后,利用k均值方法聚类图像中的角点,自适应地调整正常行为角点动能,定义每一类的局部异常程度为角点平均动能与正常时的比值,整体运动异常程度为局部异常程度之和;最后,如果整体异常程度大于异常阈值为异常行为,否则为正常行为.实验结果表明:该方法能够检测出多种群体异常行为且实时性强于Harris、SIFT(Scale-Invariant Feature Transform)和SURF(Speed Up Robust Features)角点,精确度高于光流法、社会力法和图分析法.
In order to improve the performance of real-time and detection accuracy, this paper presents a method to detect abnormal crowd behavior using corner kinetic. First, the optical flow of FAST corners is calculated using Pyramid Lucas-Kanade optical flow method and the moving corners are selected. Then, the corners are clustered using k-means method and the normal corner kinetic adaptively. The local abnormal degree is defined as the ratio of the average kinetic energy of the corner in each class with the normal. The global abnormal degree is the sum of the local abnormal degrees. Finally, it is believed as abnormal behavior if the global abnormal degree is greater than the threshold; otherwise, it is considered as normal behavior. Experimental results show that the method can detect different abnormal behavior with higher real-time performance than Harris corners, SIFT and SURF and higher detection accuracy than optical flow method, social force model and graph analysis method.
出处
《郑州大学学报（工学版）》
CAS
北大核心
2015年第3期20-24,共5页
Journal of Zhengzhou University（Engineering Science）
基金
国家自然科学基金资助项目(61172152)
关键词
群体异常行为
K均值
角点动能
异常程度
自适应
abnormal crowd behavior
k-means
corner kinetic
abnormal degree
adaptive
