基于特征融合的视频异常事件检测方法  

Video Abnormal Event Detection Method Based on Feature Fusion

作　　者：姚明海[1] 王娜 林英建[1] YAO Minghai WANG Na LIN Yingjianl(Department of College Foundation Education, Bohai University, Jinzhou 121013, China Department of Computer Science, Jinzhou Teacther＇s Training College, Jinzhou 121000, China)

机构地区：[1]渤海大学大学基础教研部,辽宁锦州121013 [2]锦州师范高等专科学校计算机系,辽宁锦州121000

出　　处：《吉林大学学报（信息科学版）》2016年第3期441-448,共8页Journal of Jilin University（Information Science Edition）

基　　金：辽宁省教育厅科学技术研究基金资助项目(L2014450);辽宁省社会科学规划基金资助项目(L13BXW006;L13BXW013)

摘　　要：为有效对视频数据进行降维并去除特征集合中的冗余信息,以提高异常事件的检测效率,从特征提取和选择的角度提出了融合特征区分度和相关性的视频异常事件检测方法。利用视频数据的时空邻域信息进行特征提取。通过分析特征的判别力和相关性进行特征选择,从而去除特征集合中的冗余信息,提高异常事件检测的效率和准确性。实验结果表明,该方法的检测准确率都优于其他传统方法,能有效地对场景中发生异常事件的区域进行准确定位。In order to effectively reduce the data dimension and remove the information of video data and improvethe efficiency of abnormal event detection, the video abnormal event detection method based on the distinctionand correlation is proposed. The method extracts features of the video data by analyzing the temporal and spatialneighborhood information of data, removes redundant information in the feature set and improves the efficiency ofabnormal event detection by analysis of the distinction and correlation of feature. The method is compared withthe traditional method in the simulation. The experimental results show that the detection accuracy of videoabnormal event detection method based on feature fusion is higher than other methods, the method can accuratelylocate the abnormal area in the scene.

关 键 词：特征融合 异常事件 特征选择 相关性分析 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]