Multi-view behavior identification method based on largest-interval meaning clustering

Abstract

Each embodiment mode of the invention provides a multi-view behavior identification method based on largest-interval meaning clustering. The method comprises steps that: global profile flow characteristics of each frame of each motion video sample are firstly extracted, a words-bag model is utilized to acquire characteristic expression vectors of the corresponding motion video sample frames; dimensionality reduction for the characteristic expression vectors is carried out by utilizing multi-time random mapping; the characteristic expression vectors which are acquired through dimensionality reduction at each random mapping are clustered through a largest-interval meaning clustering method; source object samples of all the samples are determined by utilizing a trained model, and separation characteristics of the source samples are directly taken as separation characteristics of an object domain; the separation characteristics of an object domain are taken as input, and a classification model is trained by utilizing a support vector machine; separation characteristics of a test sample of the object domain are acquired by utilizing the classification model of the support vector machine, and classification is carried out by utilizing a nearest neighbor classifier.

Description

technical field [0001] The invention belongs to the technical field of intelligent video monitoring, and in particular relates to a multi-view behavior recognition method based on semantic maximum interval clustering. Background technique [0002] Behavior recognition plays an important role in video surveillance. It can recognize the behavior of human body in the video, and contribute to dangerous behavior alarm and specific behavior recognition. The simplest and most effective method in behavior recognition is the method based on the bag of words model. This method first extracts the features of the video, then clusters all the features, and then performs a histogram according to the frequency of each video feature in the cluster center. graphic. But a shortcoming of this method is that it does not take spatio-temporal characteristics into account. Zhang et al. used a semantic-based linear coding method to not only consider the spatio-temporal relationship between feat...

AI Technical Summary

Problems solved by technology

However, when the shooting angle of the action changes, the accuracy of these methods will drop

This is because even actions of the same class look different when taken from different viewpoints, so accuracy drops when trained from one viewpoint and tested from another.

The intuitive method is to train in every view, but this requires enough samples, which is not feasible in practical applications.

Images