Interactive behavior analysis method based on self-adaptive order

Abstract

The invention discloses an interactive behavior analysis method based on a self-adaptive order. The method comprises the following steps: step 1, extracting and representing video features; step 2, modeling a video sequence by using a layered Dirichlet process model, detecting behaviors contained in a scene, and classifying by using a K-means algorithm to realize multivariable time sequence extraction and classification; step 3, selecting the order of the multivariate autoregression model; step 4, performing inter-group interaction behavior analysis of multi-vector process causal measurement based on the frequency domain; step 5, performing intra-group interaction behavior analysis of causal measurement of the multi-scalar process based on the frequency domain; and step 6, constructing a global and local behavior interaction network based on inter-group and intra-group causal measurements. The method has the advantages that global and local interaction behaviors in a scene can be detected, and a global and local behavior interaction network can be constructed.

Description

technical field [0001] The invention relates to the technical field of computer video processing, in particular to an interactive behavior analysis method based on an adaptive order. Background technique [0002] Video surveillance is usually aimed at public places with a large flow of people, traffic, and high density, such as subways, squares, and public transportation intersections. Recognizing the behavior pattern in the scene is an important problem in intelligent video surveillance. The purpose is to detect multiple behaviors as much as possible with an unsupervised method and establish the interactive relationship between them. Usually the behavior in the video can be regarded as multiple time variables, so analyzing the interaction between behaviors is to analyze the interaction between multivariate time series data. As a powerful tool for time series data analysis, Granger causality measures based on multivariate autoregressive play an important role in understandi...

AI Technical Summary

Problems solved by technology

(See: Siggiridou E, Kugiumtzis D. GrangerCausality in Multivariate Time Series Using a Time-Ordered Restricted Vector Autoregressive Model [J]. IEEE Transactions on Signal Processing, 2016, 64(7): 1759-1773.) The above two methods are essentially Both use dimensionality reduction to realize the detection of group structure in multivariate, but it is difficult to realize global and local interactive behavior analysis at the same time

Images