Two-person interaction recognition method based on knowledge embedding graph convolutional network

Abstract

The invention discloses a two-person interaction recognition method based on a knowledge-embedded graph convolutional network, which mainly solves the problems that the existing technology ignores the two-person correlation and cannot extract the two-person interaction features, and the two-person interaction recognition accuracy rate is poor. The implementation plan is as follows: 1) Design the knowledge given graph to establish the direct connection between the skeleton points of the interactive pair; 2) Design the knowledge learning graph and adaptively establish the connection between the skeleton points; Complementary knowledge embedding graph convolution blocks; 4) Connect 10 knowledge embedding graph convolution blocks, a global pooling layer and a fully connected layer in sequence to form a knowledge embedding graph convolutional network; 5) Training knowledge embedding graph The convolutional network obtains the optimal model; 6) Input the two-person skeleton sequence into the optimal model to identify the two-person interaction behavior. The invention improves the accuracy rate of two-person interactive behavior recognition, and can be used for video retrieval, human-computer interaction and video understanding.

Description

technical field [0001] The invention belongs to the technical field of video processing, and further relates to a two-person interactive recognition method, which can be used for video retrieval, human-computer interaction and video understanding. Background technique [0002] Two-person interaction recognition can utilize many different modalities, such as red-green-blue color mode image RGB, depth and skeleton. Compared with the RGB and depth modes, the skeleton mode has the characteristics of less storage capacity and strong robustness to changes in environmental factors, and the skeleton data is a high abstraction of human characteristics, which can be well suited for two-person interaction identify. The current two-person behavior recognition methods for skeletal modalities are mainly based on recurrent neural networks, convolutional neural networks, and graph convolutional networks. Based on the method of recurrent neural network and convolutional neural network, the...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to propose a two-person interaction recognition method based on knowledge-embedded graph convolutional network, to solve the existing method of constructing a skeleton graph according to the natural connection of the human body, ignoring the correlation of the skeleton points between the two interacting people, which cannot be guaranteed Suitable for all two-person interactive behaviors, and the problem of poor recognition effect

Images