Action recognition method based on unsupervised graph sequence predictive coding and storage medium

Abstract

The invention relates to an action recognition method based on unsupervised graph sequence predictive coding and a storage medium, the action recognition method comprises model training and use, is used for recognizing various actions performed by a human body in a skeleton sequence, and aims to solve the problems that an existing action recognition method highly depends on a large amount of labeled data. The precision is low under the condition that only a small amount of labels exist, and an existing unsupervised method does not utilize topological information of a graph in overfitting and is poor in serious generalization ability. The method of the system comprises: carrying out view angle invariant transformation, resampling and block-level skeleton graph data enhancement on skeleton sequence data; enabling embedding of space-time diagram convolution skeleton sequence block to express extraction; aggregating context features by the graph convolutional recurrent neural network; constructing positive and negative sample pairs through predictive coding; and extracting features through the pre-training model, and obtaining an action category corresponding to the to-be-recognized skeleton sequence by using the classifier. Compared with the prior art, the method has the advantages of low training difficulty, high recognition precision, excellent performance and the like.

Description

technical field [0001] The invention relates to the technical field of action recognition, in particular to an action recognition method and storage medium based on unsupervised graph sequence predictive coding. Background technique [0002] Among computer vision tasks, action recognition is a hot topic that has attracted much attention now. Unmanned robots, smart cities, intelligent transportation and other fields all need to analyze and recognize the behavior of the human body. In recent years, with the attention and utilization of graph convolution by more and more researchers, the development of pose estimation algorithms and depth sensors, as well as the robustness of skeleton data and the characteristics of focusing on the action itself without visual features, using skeleton Action recognition from data has become a hot research topic at present. [0003] Early action recognition was mainly based on static images. In recent years, with the deepening of research, mo...

AI Technical Summary

Problems solved by technology

However, these encoder-decoder models usually flatten the spatial channels into a single feature vector, ignoring the spatial relationship of the skeleton map

And these excuse tasks often suffer from overfitting problems, and it doesn't always help in downstream tasks

Images