Action recognition model training method, and action recognition method and device

Abstract

The invention provides an action recognition model training method, and an action recognition method and device. The training method comprises the steps of obtaining node data of all nodes in a presetaction node set; calculating a connection edge set of all nodes in the node set and edge data of each connection edge according to the node data; constructing a graph structure of action data according to the node data and the edge data of the connecting edge; and taking the graph structure of the action data as model input, taking an expression recognition classification result as model output,and carrying out supervised training on a preset graph convolutional neural network expression recognition model, wherein the edge data of the connecting edge in the graph structure of the action datais used as model input. The action data is recorded and calculated based on the graph structure, the effective data volume participating in deep learning is further improved, better recognition precision can be obtained, and dependence on sample data precision is reduced.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence, in particular to an action recognition model training method and an action recognition method and device. Background technique [0002] At present, human body action recognition, especially expression recognition, is mainly based on the time series of actions or facial expressions in collected images or videos, and the classification of actions or expressions is realized through deep neural networks, especially deep convolutional neural networks. However, based on images and video sequences to recognize expressions and actions, especially micro-expressions and micro-actions, convolutional neural network algorithms are often limited by the accuracy of expression capture, and it is difficult to have high accuracy. Among them, micro-expression classification is more difficult than macro-expression (emotion classification), mainly because the data is sparse, it needs to be loc...

AI Technical Summary

Problems solved by technology

Since the extracted node information from the image sequence often only has the characteristics of one aspect of the pixel (the R, G, B information of the pixel is regarded as the same feature aspect), so that the action recognition of the graph neural network based on the feature point as the node, especially It is an expression recognition method, especially limited by the accuracy of expression capture and node selection. When the accuracy is not good or the node selection is inaccurate, ideal recognition results are often not obtained.

Images