Crowd behavior identification method

Abstract

The invention provides a crowd behavior identification method. The crowd behavior identification method comprises the steps that a movement track contained in a predefined movement mode is divided into a data point set, the data point set serves as a training sample of a Bayes classifier, and training is carried out through the Bayes classifier to obtain a training model; crowd movement information is obtained from a video sequence; traversal is carried out on the crowd movement information through a sliding window to obtain a movement track, the movement track is expressed through discrete track points, and the track points are clustered to obtain a detection movement mode; the detection movement mode is divided into a track point set, and the track point set serves as a test set to be input into the training model; a result of matching between the test set and the predefined movement mode is fed back by the training model, if matching is successful, crowd behaviors are determined as normal behaviors, and if matching fails, the crowd behaviors are determined as abnormal behaviors.

Description

technical field [0001] The invention belongs to the technical field of digital image processing, in particular to a crowd behavior recognition method. Background technique [0002] Crowd behavior detection and recognition is an important part of urban security and intelligent video surveillance, and it is also a basic requirement for autonomous crowd management and regulation. In surveillance video analysis, the current behavior detection and recognition is mainly aimed at the detection and recognition of the behavior of a single person and the behavior of a small number of people. Due to the characteristics of high density and complex and diverse movement conditions of the crowd, there are certain limitations in the actual analysis. Difficulty. At present, the most commonly used method is to build a model for crowd targets, and use the model to complete crowd behavior recognition and crowd anomaly detection. [0003] The current research method for the crowd target studies...

AI Technical Summary

Problems solved by technology

Generally, the crowd model is obtained through a large amount of training, but due to the complex and changeable motion in the video, it is extremely difficult to establish a crowd model with strong adaptability for the video sequence, and it is difficult to be widely used in actual scenes.

[0005] Therefore, in practical applications, the existing technology is poorly adaptable to different application scenarios

When the scene changes, it is often necessary to retrain the model, which will significantly increase the cost and development difficulty

Images