Abnormal Activity Detection Using Spatio-Temporal Feature and Laplacian Sparse Representation

Abnormal activity detection in a video is a challenging and attractive task. In this paper, an approach using spatio-temporal feature and Laplacian sparse representation is proposed to tackle this problem. To detect the abnormal activity, we first detect interest points of a query video in the spatio-temporal domain. Then normalized combinational vectors, named HNF, are computed around the detected space-time interest points to characterize the video. After that, we utilize the Laplacian sparse representation framework and maximum pooling method to gain a more discriminative feature vector from the HNF set. Finally, the support vector machine (SVM) is adopted to classify the feature vector as normal or abnormal. Experiments on two datasets demonstrate the satisfactory performance of the proposed approach.