Online least squares one-class support vector machines-based abnormal visual event detection

Abstract

The abnormal event detection problem is an important subject in real-time video surveillance. In this paper, we propose a novel online one-class classification algorithm, online least squares one-class support vector machine (online LS-OC-SVM), combined with its sparsified version (sparse online LS-OC-SVM). LS-OC-SVM extracts a hyperplane as an optimal description of training objects in a regularized least squares sense. The online LS-OC-SVM learns a training set with a limited number of samples to provide a basic normal model, then updates the model through remaining data. In the sparse online scheme, the model complexity is controlled by the coherence criterion. The online LS-OC-SVM is adopted to handle the abnormal event detection problem. Each frame of the video is characterized by the covariance matrix descriptor encoding the moving information, then is classified into a normal or an abnormal frame. Experiments are conducted, on a two-dimensional synthetic distribution dataset and a benchmark video surveillance dataset, to demonstrate the promising results of the proposed online LS-OC-SVM method.

Figure 1.

Examples of the normal and abnormal scenes. (a) A normal lawn scene: all the people are walking; (b) An abnormal lawn scene: all the people are running.

Figure 2.

Covariance descriptor computation based on the features of a video frame.

Figure 3.

Major processing states of the proposed abnormal frame event detection method. The covariance of the frame is computed.

Figure 4.

Synthetic dataset. (a) Square; (b) ring-line-square.

Figure 5.

Offline, online LS-OC-SVM and sparse online LS-OC-SVM results of “square”. The figures might be viewed better electronically, in color and enlarged, (a) The contours when all the training data are learned as one batch offline; (b) The contours when the training data are learned via online LS-OC-SVM; (c) The blue circle (pointed out by the arrow) shows the original dictionary. The red points show the 232 new data included into the dictionary via sparse online LS-OC-SVM; (d) The contours when the training data are learned via sparse online LS-OC-SVM.

Figure 6.

Offline, online LS-OC-SVM and sparse online LS-OC-SVM results of “ring-line-square”, (a) The contours when all the training date are learned as one batch offline; (b) The contours when the training data are learned via online LS-OC-SVM; (c) The blue circle (pointed out by the arrow) shows the original dictionary. The red points show the 534 new data included into the dictionary via sparse online LS-OC-SVM; (d) The contours when the training data are learned via sparse online LS-OC-SVM.

Figure 7.

Detection results of the lawn scene, (a) The detection result of a normal frame; (b) The detection result of an abnormal panic frame; (c) Receiver operating characteristic (ROC) curve of covariance descriptors constructed from different features F of the lawn scene detection results via LS-OC-SVM. All the training samples are learned together offline. The biggest AUC value is 0.9874; (d) ROC curve of detection results via online LS-OC-SVM. The biggest AUC value is 0.9874.

Figure 8.

Detection results of the indoor scene. (a) The detection result of a normal frame; (b) The detection result of an abnormal panic frame; (c) ROC curve of covariance descriptors constructed from different features F of the indoor scene results via LS-OC-SVM. All the training samples are learned together offline. The biggest AUC value is 0.8900; (d) ROC curve of detection results via online LS-OC-SVM. The biggest AUC value is 0.8904.

Figure 9.

Detection results of the plaza scene. (a) The detection result of a normal frame; (b) The detection result of an abnormal panic frame; (c) ROC curve of covariance descriptors constructed from different features F of the plaza scene results via LS-OC-SVM. All the training samples are learned together offline. The biggest AUC value is 0.9800; (d) ROC curve of detection results via online LS-OC-SVM. The biggest AUC value is 0.9839.

Figure 10.

ROC curve of University of Minnesota (UMN) dataset. (a) Sparse online LS-OC-SVM results in the lawn scene. The biggest AUC value is 0.9510; (b) Sparse online LS-OC-SVM results in the indoor scene. The biggest AUC value is 0.8886; (c) Sparse online LS-OC-SVM results in the plaza scene. The biggest AUC value is 0.9515; (d) The ROC curve of the best performance of the lawn, plaza and indoor scene when the training samples are learned via LS-OC-SVM offline. The biggest AUC values of the lawn, indoor and plaza scene are 0.9874, 0.8900 and 0.9800.