Video monitoring crowd behavior identification method based on deep residual neural network convolution

Abstract

The invention discloses a video monitoring crowd behavior recognition method based on deep residual neural network convolution, and the method comprises the steps: inputting a data source into a residual neural network for image processing, and dividing the data source into different types of sub-crowd behavior features; creating a subclass in each sub-crowd behavior feature by using PCA, calculating a dispersion matrix in the subclass to extract a key subclass, judging abnormal conditions of all the subclasses, and outputting a safety index in combination with crowd behaviors of the subclasses; enabling the detection module to find corresponding frame images extracted from crowd subclasses with low safety indexes, and detect whether the video frame images have abnormal behaviors or not through a background difference method and a regional optical flow method; finding similarity measurement between the frame images of the two different behavior groups on cosine distance measurement soas to track dynamic videos of abnormal crowd behaviors. According to the method, useful crowd behavior category information is effectively extracted, and key decisions and supports can be provided forpublic safety and crowd management in time.

Description

technical field [0001] The invention relates to the technical field of crowd behavior recognition and analysis in surveillance video, in particular, to a video surveillance crowd behavior recognition method based on deep residual neural network convolution. Background technique [0002] It is extremely difficult for human observers to monitor large numbers of individuals, their behavior and activity from large camera topologies. Areas with behavioral anomalies are usually highly congested urban areas, and the intelligent detection of dense crowd behavior to extract useful behavior pattern information has become critical for public safety, security, crowd management, and timely provision of key decisions and support. [0003] Existing research has mainly focused on sparse and mostly staged scenes, and relatively little effort has been devoted to reliably classifying and understanding human activities in real and very crowded scenes. In overcrowded scenes, the detected target...

AI Technical Summary

Problems solved by technology

The second is an object-based approach, where each individual (person and / or object) is detected and segmented to perform motion and / or behavior analysis, which enables complex segmentation of individuals in crowded videos and tracking is a very challenging task

Images