Multi-view video anomaly detection method based on sparse coding

Abstract

The invention relates to the technical field of computer vision, in particular to a multi-view video anomaly detection method based on sparse coding. The method comprises: performing multi-view feature extraction on frame images; sparse coding being applied to the features from different perspectives to obtain sparse representations of the features from different perspectives; obtaining a consistency representation matrix under a frame image according to the sparse representation information and assigning corresponding weight values to the consistency representation matrix between two adjacentframes to obtain a dictionary A, then the reconstruction error of sparse representation coefficients being obtained by using dictionary A to test the video data of abnormal events, and the standardized multi-view video anomaly detection model being obtained. The invention extracts the multi-view angle characteristic of the video frame image, establishes the multi-view angle video anomaly detection model, integrates the characteristic information under the multi-view angle of the video to carry out the anomaly detection, and utilizes the time want-to-dry property between two adjacent frames ofthe video, reduces the loss of local information, and improves the anomaly detection accuracy.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a multi-view video anomaly detection method based on sparse coding. Background technique [0002] Anomaly detection has a lot of research in the field of computer vision. Because it has many potential applications in video surveillance, activity recognition, and scene understanding. An anomaly detection system can greatly reduce manual labor and time. However, anomalous event detection is still a very challenging task because anomalous events have no clear boundary definition. In real applications, on the one hand, abnormal events are rare compared to normal events, and we need to spend a lot of money to collect them; on the other hand, it is impossible to collect all abnormal events. Therefore, for typical anomaly detection datasets, only general scenarios are given in the training set. To identify whether anomalous events occur, the usual approach is to exploit regu...

AI Technical Summary

Problems solved by technology

In addition, the temporal correlation between two adjacent frames is often ignored by us. Some studies have shown that in sparse coding, similar features may be coded into dissimilar codes, resulting in the loss of local information.

Images