Bimodal iterative denoising anomaly detection method based on video weak mark, and terminal

Abstract

The invention provides a bimodal iterative denoising anomaly detection method and terminal based on a video weak mark, and the method comprises the steps: enabling the weak mark to serve as the noiseof an accurate label, and respectively carrying out the label denoising from an image control and a feature space; learning characteristics of normal and abnormal videos from an image space by using an auto-encoder; learning features of the video clip at different times by using a graph convolution model; and alternately updating the classifier and the de-noiser by utilizing iteration. According to the method, the weak marking problem of the video is fully considered, the difficulty of marking the data is overcome by utilizing a denoising model method, and the method has very strong robustness, can well solve the problem of marking of the video and has universal applicability in the research field that the data are difficult to collect in abnormal detection.

Description

Technical field [0001] The invention belongs to the technical field of computer vision, and is specifically a dual-modal iterative denoising abnormal detection method and terminal based on video weak markers, and particularly relates to abnormal detection focusing on abnormal crowd behavior under a surveillance camera. Background technique [0002] Surveillance cameras are increasingly being used in public places, such as streets, intersections, banks, shopping malls and other places with dense traffic. However, the ability of the relevant administrative law enforcement agencies to detect abnormal situations in surveillance video has not kept up, resulting in the inability to make full use of the resources of surveillance cameras, and there are obvious defects in their use. It is also very unrealistic to let people observe surveillance video in real time, because the number of surveillance cameras in our country is already very large. It is not only costly to rely on human resour...

AI Technical Summary

Problems solved by technology

[0007] Aiming at the weak label problem of the current data set, the present invention regards the weak label as the noise of the precise label, and denoises the label from the image control and feature space respectively; uses the self-encoder to learn the characteristics of normal and abnormal videos from the image space; uses the graph The convolutional model learns the features of video clips at different times; iteratively updates the classifier and denoiser alternately

The present invention overcomes the difficulty of labeling data by fully considering the problem of weak labeling of video, and uses the method of denoising model. Facing the research field of abnormal detection, which is difficult to collect data, it has strong robustness and can solve the problem of video very well. If labeling is a problem, it has general applicability

Images