Static gesture recognition method based on attention mechanism

Abstract

The invention discloses a static gesture recognition method based on an attention mechanism. Gesture image acquisition and size normalization are carried out; inputting the original gesture image into a first layer of a reference network; respectively inputting the first-order feature vector into a gesture detail attention module and a second layer of the reference network, fusing the second-order feature vector and the output of the gesture detail attention module, and inputting the fused second-order feature vector and the output of the gesture detail attention module into a first channel attention module; the output of the first channel attention module is input into the third layer of the reference network; inputting the first-order feature vector and the output of the channel attention module into a gesture subject attention module; fusing the output of the gesture subject attention module and the third-order feature vector, and inputting the fused output and third-order feature vector into a second channel attention module; and the second channel attention module inputs the remaining reference network and the softmax classifier to obtain a classification result of the gesture image. The method is designed according to the characteristics of static gesture recognition, and the problem that the model accuracy is not high under the condition of few training sets can be solved.

Description

technical field [0001] The invention relates to a method for extracting gesture images, in particular to a method for recognizing static gestures based on an attention mechanism. Background technique [0002] With the maturity of neural network technology, gesture recognition based on computer vision is becoming a craze. In practical applications, how to recognize gestures and the required accuracy also make real-time gesture recognition more difficult. Although gesture recognition technology has made great progress, there are still many challenges in the real environment, such as lighting, distance and many other factors will affect the performance of gesture recognition. Moreover, there are relatively few static sign language data sets disclosed at present, so there may be situations in which model training needs to be performed with few training sets. At present, gesture recognition methods based on deep learning have gradually become the mainstream. Generally, convolut...

AI Technical Summary

Problems solved by technology

[0003] Aiming at the problem of low accuracy of gesture feature extraction in the network training process under the condition of few training sets, the present invention proposes a simple and effective feature optimization module, which can specifically solve the problem of low model accuracy under the condition of few training sets The problem

Images