Group positioning and abnormal behavior detection method in video

Abstract

The invention discloses a group abnormal behavior detection algorithm in a video. Firstly, a large amount of video image data is acquired as a training sample for analyzing and identifying groups anddetecting abnormal behaviors; secondly, a crowd density estimation model is trained by adopting a neural network based on hole convolution to obtain a video image crowd density map, and point clustering is performed on the density map in combination with a clustering method to obtain the position and the size of a group; thirdly, for all the anomaly detection video data sets, a feature extractionnetwork is used for extracting spatial and temporal features of the anomaly detection video data sets, input of a training neural network is obtained, training samples are input into a full-connectionneural network with set parameters, the neural network is trained until cost loss is reduced to a certain degree and the maximum number of iterations is achieved, and a trained model is obtained; andfinally, group information obtained by group identification is taken as a region of interest, spatial and temporal features of the test video are extracted, and the spatial and temporal features areinput into the trained anomaly detection model to obtain an anomaly detection score of the video.

Description

technical field [0001] The present invention relates to the field of video image processing and video analysis, in particular, the present invention relates to group identification in video and several types of specific abnormal behavior detection methods. Background technique [0002] Group abnormal behavior analysis in video is of great significance in intelligent surveillance systems and UAV aerial video processing, and has broad application prospects. Anomaly detection is a research hotspot in the field of computer vision, and it is also a difficult point. How to effectively extract the required information from a large amount of video data, and timely alarm or even early warning of abnormal behavior will have a major impact on the field of public security, not only saving a lot of manpower and material resources, but also maximizing the protection of people's safety. Safety of life and property. However, most of the current video surveillance systems need to manually ...

AI Technical Summary

Problems solved by technology

This kind of method theory can get better results, but it is difficult to obtain a large number of abnormal behavior data sets, and manual labeling is also time-consuming and laborious

Images