Video anomaly detection method based on weighted tensor subspace background modeling

Abstract

The invention discloses a video anomaly detection method based on weighted tensor subspace background modeling, which is mainly used for solving the problem that the prior art can not filter out outliers in an image due to the ignorance of space structure information of the image. The implementation process is as follows: firstly regarding training data and observation data as two-order tensors, adopting the tensor analysis method to calculate a projection matrix on each mode, and constituting a tensor subspace; then carrying out robustness analysis on the observation data, weighting each element in the observation data, updating the tensor subspace, projecting the weighted observation data onto the subspace, and reconstructing a background image; and finally carrying out similarity measurement on a reference background and the reconstructed background, and detecting whether an anomalous event happens in a video scene or not. Compared with the prior art, the method can keep the space structure information of the image, filter out the outliers in the image and have robustness. The method can be used for anomalous event detection under the conditions of fixed scenes and slowly-changing illumination in the fields of security protection and monitoring.

Description

technical field [0001] The invention relates to the technical field of video monitoring, in particular to a video anomaly detection method, which can be used for abnormal event detection under fixed scenes and slowly changing illumination conditions in the field of security and monitoring. Background technique [0002] In the field of public security, video surveillance is playing an increasingly important role. Intelligent video surveillance is reflected in that it can automatically identify abnormal events through image analysis, reduce the workload of security monitoring personnel, and reduce missed and false positives of abnormal events. Video anomaly event detection integrates computer vision, image processing, pattern recognition and other multi-disciplinary technologies, which has important scientific significance and broad application prospects. In recent years, with the rapid growth of population and the increasingly complex urban environment, various crimes and te...

AI Technical Summary

Problems solved by technology

For example, the literature Y.Li.On incremental and robust subspace learning.Pattern recognition, 37:1509-1518, 2004. However, this method is based on vector processing, that is, the image is regarded as a high-dimensional vector, and the spatial structure information of the image is ignored. , resulting in the inability to correctly judge some of the outlier points in the image, and unable to filter out these outlier points. These outlier points are learned into the background model, causing a large change in the background model and affecting the reconstruction quality of the background model.

Images