Method and system for video abnormal event detection based on atom feature bag model

Abstract

The present invention provides a method and system for video abnormal event detection based on an atom feature bag model. The method comprises: performing video sampling, dividing a video into a plurality of space-time bodies taken as video events, dividing each space-time body into time-space blocks, extracting GCM descriptors from the time-space blocks and STCV descriptors between adjacent time-space blocks, using a BoAF model to obtain a BoAF representation of each video event, employing the K-SVD algorithm to learn an over-complete dictionary of a BoAF representation of a normal event, calculating sparse reconstruction cost of a BoAF representation of each event in the over-complete dictionary, and taking a video event having a sparse reconstruction cost larger than an experience threshold as an abnormal event. The method and system provided by the invention can detect abnormal behaviors in a scene, can effectively detect abnormal events caused by an event structure context, and can obtain a higher detection rate in a complex and crowded scene.

Description

technical field [0001] The invention relates to a technology in the field of image processing and recognition, in particular to a method and system for detecting abnormal video events based on an atomic feature bag model. Background technique [0002] Traditional monitoring methods rely on manpower, which is usually inefficient and difficult to process the massive data generated by cameras. Intelligent video surveillance is an urgent need under the current situation, and it can make up for the shortage of manpower. [0003] Abnormal video events refer to the events caused by the monitoring target that do not conform to the event rules in the scene and are potentially dangerous. Intelligent video surveillance can detect abnormal events in the video scene in a timely manner and issue an alarm to remind personnel to deal with them, and can also accurately locate the monitoring target that caused the abnormal event. [0004] The bag-of-words (BoW) model that is commonly used n...

AI Technical Summary

Problems solved by technology

However, the existing technology will generate a large number of meaningless feature points in the presence of dynamics, such as trees, fluctuating water surfaces, etc., and crowded scenes, and the spatiotemporal feature descriptors and bag-of-words models used cannot reflect the relationship between local feature descriptors. It is difficult to detect the abnormal events caused by the abnormal changes of the spatio-temporal combination relationship, ignoring the intra-class differences between the same type of feature descriptors, resulting in large approximation errors and resulting in missed detection or false detection of abnormal events. check

Images