Two-stage behavior recognition fine classification method based on graph convolutional network

Abstract

The invention discloses a two-stage behavior recognition fine classification method based on a graph convolutional network. According to the method, the accuracy of behavior recognition is improved mainly by reclassifying a difficult category set, and the method is divided into three stages: in the first stage, training a coarse classification model; in the second stage, acquiring difficult category sets and training difficult category set models, acquiring the difficult category sets by a confusion matrix of the rough classification model on the test set and a union-check set algorithm, and training the difficult category set models for the different difficult category sets; and in a third stage: during on-line inference, according to an inference result of the rough classification model, inputting the samples which need to be finely classified into the difficult category set model for re-classification. According to the method provided by the invention, aiming at the problem that a model is difficult to classify similar actions, the problem that the similar actions are difficult to classify is relieved to a certain extent by a rough classification-fine classification two-stage architecture, the accuracy of behavior recognition is improved, and a better result is obtained on a public data set.

Description

Technical field [0001] The present invention is in the field of behavior recognition, in particular to a method of identifying a fine classification based on the behavior of the two-stage network of FIG convolution. This method is superior to conventional methods for the single-flow model to identify a single popular classification task. Background technique [0002] Behavior recognition is a hot research tasks in computer vision: behavior recognition goal is to identify the behavior of the human image or video. Behavior recognition method based on the depth of learning methods can be divided based on RNN, CNN-based approach, GCN-based approach. GCN based method because of its high accuracy, many researchers to be selected, the method is employed in the present invention. At present, the main methods are mostly focused on improving the overall classification accuracy of behavior recognition model, and most of the behavior recognition model for class actions in the action part of ...

AI Technical Summary

Problems solved by technology

This method is an improvement on the defect that the existing model does not classify highly similar actions well.

Images