Gesture recognition method and system based on knowledge distillation and attention mechanism

Abstract

The invention relates to a gesture recognition method and system based on knowledge distillation and an attention mechanism, and the method comprises the steps: obtaining a gesture picture data set, inputting the data set into a teacher model of a three-level network, carrying out the gesture recognition of the gesture picture data set through the teacher model, and outputting a gesture classification result. then, carrying out training with knowledge distillation on the student model of the secondary network through a manual annotation data set, a random data set automatically generated by a computer and a gesture classification result of the teacher network with pictures not marked, and obtaining a trained secondary network student model; and finally, acquiring a gesture picture data set, inputting the data set into the trained student model of the secondary network, identifying gestures through the trained student model, and outputting to obtain a gesture classification result. According to the invention, the gesture recognition precision and speed are improved, and the mobile terminal deployment of the model is realized.

Description

technical field [0001] The present application relates to the field of computers, in particular to a gesture recognition method and system based on knowledge distillation and attention mechanism. Background technique [0002] Gesture recognition usually uses traditional machine learning methods such as SVM to directly classify gesture pictures. However, due to the lack of recognized gesture types, it is necessary to train the model from scratch to expand the gesture types. In order to facilitate the expansion of gesture types, the hand key point detection can be performed first, and then the hand key point gesture recognition can be performed. However, the hand key point detection model constructed by the deep neural network is often very large, which is not conducive to mobile terminal deployment and rapid inference. The model needs to be compressed; in addition, when performing gesture recognition after the key point detection of the hand, the accuracy of gesture classific...

AI Technical Summary

Problems solved by technology

[0004] The embodiment of the present application provides a gesture recognition method and system based on knowledge distillation and attention mechanism, so as to at least solve the problems of low gesture recognition accuracy, slow inference speed and limited application scenarios in the related art, which are not conducive to mobile terminal Deployment issues

Images