Human body behavior identification method based on deep recursive and hierarchical condition random fields

Abstract

The invention discloses a human body behavior identification method based on deep recursive and hierarchical condition random fields. The human body behavior identification method comprises the following steps: firstly, independently extracting the body gesture of a behavior act body and object information which may mutually interact with the body gesture in the RGB-D video of a behavior act scene shot by a RGB-D camera, taking the two types of information as an intermediate layer state of the deep recursive and hierarchical condition random field, modeling correlation between a current state and all current generated prediction output state sets in a prediction output target state layer, and constructing the deep recursive and hierarchical condition random field model; secondly, adopting a structured support vector machine driven by a BCFW (Block-Coordinate Primal-Dual Frank-Wolfe) optimization method to learn a discrimination classification model based on a human body behavior sequence; and finally, according to the model parameter obtained by learning and the obtained discrimination model, predicting the category of the human body behavior sequence to be tested. The human body behavior identification method exhibits obvious robustness and improves the identification accuracy of human body behavior act to a certain degree.

Description

technical field [0001] The invention relates to a human behavior recognition method, in particular to a human behavior recognition method based on Deep Recursive and Hierarchical Conditional Random Fields (DR-HCRFs), which belongs to the technical field of computer vision behavior recognition. Background technique [0002] Human behavior recognition plays an important role in computer vision, and has a wide range of applications in the fields of intelligent monitoring, human-computer interaction and sports video processing. [0003] In recent years, research on behavior recognition in indoor scenes mainly uses the method of probabilistic graphical model to classify and analyze human behavior. Common probabilistic graphical models are mainly divided into two structures: generative models and decision models. Common generative models are: Hidden Markov Model (HiddenMarkovModel), Bayesian Networks (DBNs), Semi-Markov Models (Semi-MarkovModels). Generative models need to model...

AI Technical Summary

Problems solved by technology

[0005] The above-mentioned existing behavior recognition methods based on probabilistic graphical models have not considered the internal representation of the target state and the high-order correlation between states, and still have the problem of low recognition accuracy

Images