东篱科研大数据发现系统（DRDS）

位置：成果数据库 > 期刊 > 期刊详情页

基于多示例学习的异常行为检测方法

ISSN号：1003-6059
期刊名称：《模式识别与人工智能》
时间：0
分类：TP181[自动化与计算机技术—控制科学与工程;自动化与计算机技术—控制理论与控制工程]
作者机构：[1]南京大学计算机科学与技术系计算机软件新技术国家重点实验室,南京210093
相关基金：国家自然科学基金项目（No.61035003,61175042,60721002）、国家973计划项目（No.2009CB320700）、科技部国际合作专项项目（No.2010DFA110307）、教育部新世纪人才支持计划项目（No.NCEF-10-0476）和江苏省社会发展项目（No.BE2010638）资助

关键词：异常行为检测, 轨迹分段, 层次狄利克雷过程-隐马尔科夫模型(HDP—HMM), 多示例学习, Abnormal Event Detection, Trajectory Hidden Markov Model （HDP-HMM） Segmentation, Hierarchical Dirichlet Process- Multi-Instance Learning

中文摘要：

在基于轨迹分析的异常行为检测方法中，被标记为异常的轨迹往往仅在整条轨迹的某个局部存在异常，轨迹的其余部分都是正常行为．然而，传统的基于整条轨迹建模的方法很难检测轨迹的局部异常．针对上述问题，提出一种在多示例学习框架下基于轨迹分段的异常行为检测方法．该方法首先根据轨迹的曲率，将轨迹分割成若干相互独立的子段．然后采用层次狄利克雷过程一隐马尔科夫模型对每个子段建模．最后在多示例学习框架下，以整条轨迹为包，正常轨迹为负包，异常轨迹为正包，轨迹子段为包的示例进行学习．通过实验验证，该方法在准确率和召回率上都优于传统的基于轨迹建模的方法．

英文摘要：

In trajectory-based abnormal event detection, abnormal trajectory usually has abnormality in some parts of the whole trajectory and the rest are normal. However, most of the previous approaches are not able to detect this kind of abnormality easily. Aiming at the problem, an approach is proposed for abnormal event detection based on trajectory segmentation within the framework of multi-instance learning. In the proposed method, every trajectory is segmented into independent sub-trajectories based on their curvature firstly. Then, the sub-trajectories are modeled by hierarchical Dirichlet process-hidden Markov model （HDP-HMM）. Finally, within the multi-instance learning framework, the whole trajectory is considered as bags while normal ones are negative bags, abnormal ones are positive bags, and sub- trajectories are instances in the bags. Experimental results show the proposed method achieves higher precision and recall than traditional ones.