Motion Anomaly Detection Method Based on Hierarchical Independent Component Coding

Abstract

The present invention discloses a motion abnormity detection method based on level independent component coding. On the basis of obtaining the first-layer training set of an optical flow region block, a normalization matrix and feature vectors thereof are constructed, the feature vectors are subjected to orthogonalization and are taken as initial elements, a hyperbolic tangent conversion loss function is employed to perform element learning, and a motion high frequency mode is excavated as S1-layer elements; the S1-layer learning elements are employed to perform convolution of optical flow images to obtain initial response, the cut-off linearity correction is employed to obtain the C1-layer response; the C1-layer response employs spatial sampling to construct a second-layer training set, S2-layer independent component elements are subjected to learning, the cut-off linearity correction is employed to obtain C2-layer response, and S3-layer independent component elements are obtained through excavation; and the optical flow of the test video sequence is subjected to layer-to-layer convolution of S1-layer, S2-layer and S3-layer elements to obtain motion mode response, a clustering method is employed to generate a multi-clustering center, the multi-Gaussian kernel density estimation is employed to realize abnormal probability estimation to realize motion abnormity detection and region marking.

Description

technical field [0001] The invention relates to the field of anomaly detection methods, in particular to a motion anomaly detection method based on hierarchical independent component coding. Background technique [0002] In recent years, research on the analysis and understanding of video scenes has attracted the attention of many researchers in the field of computer vision, who are committed to researching new technologies and methods to analyze and understand scene content more accurately and quickly, so as to more effectively assist monitors to obtain Accurate information and handle emergencies, and minimize false positives and negatives, and play a role in supervision and management. Abnormal event detection in video scenes is one of the important research contents, and it is also a hot spot and difficult point of research. [0003] The most classic method of anomaly detection is usually based on manually designed features for anomaly detection. "A system for learning ...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to provide a motion anomaly detection method based on hierarchical independent component coding to solve the problem that the prior art anomaly representation method has insufficient ability to describe the visual perception hierarchical relationship

Images