基于运动标签的异常行为检测算法  

ANOMALOUS BEHAVIOURS DETECTION ALGORITHM BASED ON MOTION LABEL

作　　者：邹一波[1] 陈一民[1] 

机构地区：[1]上海大学计算机工程与科学学院,上海200444

出　　处：《计算机应用与软件》2015年第5期238-240,266,共4页Computer Applications and Software

基　　金：国家科技支撑计划课题(2006BAK12B10);上海市国际科技合作基金项目(12510708400)

摘　　要：针对视频监控中异常行为检测的问题,提出了一种基于运动标签的检测算法。通过对视频结构语义的分析,针对像素块的运动标签在时间维度和空间维度的分布关系,定义五种低维度的行为特征描述符作为视频中行为特征的表示。将这些提取的行为特征作为SVM分类器的输入,训练和建立了基于运动标签的异常行为检测模型。实验表明,该算法与同类算法相比,取得了更好的检测效果。For anomalous behaviours detection in video surveillance, we presented a motion label-based detection algorithm. By analysing the semantics of video structure, and aiming at the distribution relationship of pixel blocks＇ motion labels in time dimensionality and spatial dimensionality, we defined five low-dimensional behaviour feature descriptors as the representation of video behaviour feature. These extracted features are used as the input of the SVM classifier, and the anomalous behaviours detection model based on motion label is trained and built up. It is demonstrated by experiments that the algorithm achieves better detection results than similar algorithms.

关 键 词：异常行为检测 运动标签 行为描述符 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]