时空特征能够同时在时间和空间维度捕捉人体运动信息, 以三维丰富的信息量表征人体运动具有极大的优势, 本文基于时空特征提出一种新的行为表示方法进而应用于人体异常行为检测。首先利用改进的方法检测时空兴趣点并提取三维尺度不变特征变换描述子(3D SIFT), 同时提取时空兴趣点的位置分布信息(LOC)与之结合作为运动特征表示, 然后本文提出在单帧以及所有帧之间特征信息进行两次主成分分析降维处理, 大大降低特征的维数, 最后利用支持向量机算法在公开的Weizmann数据库进行异常行为检测实验并得到了较高的正确检测率, 验证了所提方法的有效性。
The spatio-temporal feature has a great advantage in capturing human motion information in both temporal and spatial scales, and in describing the human body movement with rich amount of information.A new method of motion representation based on spatio-temporal feature is proposed and applied to detect abnormal human behaviors in this paper.The interest points are detected by using an improved method, and a 3-dimensional scale-invariant feature transform (3D SIFT) descriptor is extracted, combined with the position distribution information (LOC) of the interest points.In order to enhance integrity and reduce dimension of the feature, twice Principal Component Analysis (PCA) are done on the feature of single frame and all frames.The feature is tested by using the support vector machine (SVM) algorithm on the public Weizmann dataset and high positive detection rates are reached.The results show that the feature has good robustness and applicability in effectively describing human motion information.
[1]Chianese A, Moscato V, Picariello A.Detecting abnormal activities in video sequences[C].Proceedings of the 2008 Ambi-Sys workshop on Ambient Media Delivery and Interactive Television, 2008:1-8.
[2]Jiang F, Wu Y, Katsaggelos A K.A dynamic hierarchical clustering method for trajectory-based unusual video event detection[J].IEEE Transactions on Image Processing, 2009, 18(4):907-913.
[3]周宜波, 何小海, 张生军, 等.一种新的异常行为检测算法[J].计算机工程与应用, 2012, 48(3):192-194.
[4]Zhang J, Liu Z.Detecting irregularities by image contour based on fuzzy neural network[C].Proceedings of the 3th International Conference on Innovative Computing Information and Control, 2008:401-404.
[5]Ryan D, Denman S, Fookes C, et al.Textures of optical flow for real-time anomaly detection in crowds[C].Proceedings of the 8th IEEE International Conference on Advanced Video and Signal-Based Surveillance, 2011:230-235.
[6]Ma X, Liang G, Yu W, et al.Abnormal behavior detection based on global motion orientation[C].Proceedings of the 2th International Conference on Systems Engineering and Modeling, 2013:461-464.
[7]Kratz L, Nishino K.Anomaly detection in extremely crowded scenes using spatio-temporal motion pattern models[C].Proceedings of IEEE Conf.Comput.Vision Pattern Recog, 2009:1446-1453.
[8]He L, Wang D, Wang H.Human abnormal action identification method in different scenarios[C].Proceedings of International Conference on Digital Manufacturing and Automation.2011, 594-597.
[9]Lu H, Li G, Shu G, et al.Abnormal behavior analysis using LDA[C].Proceedings of International Conference on Audio, Language and Image Processing, 2012:96-100.
[10]Dollar P, Rabaud V, et al.Behavior recognition via sparse spatio-temporal features[C].Proceedings of IEEE International Workshop on Visual Surveillance and Performance Evaluation of Tracking and Surveillance, 2005:65-72.
[11]Bregonzio M, Gong S, Xiang T.Recognizing action as clouds of space-time interest points[C].Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, 2009:1948-1955.
[12]Scovanner P, Ali S, Shah M.A 3-dimensional sift descriptor and its application to action recognition [C].Proceedings of the 15th International Conference on Multimedia, 2007:357-360.
[13]丁世飞, 齐丙娟, 谭红艳.支持向量机理论与算法研究综述[J].电子科技大学学报, 2011, 40(1):2-10.
