Application Research Of Gesture Interaction Technology Based On Lightweight Convolution Neural Network

Gesture interaction is an important means of interaction in human-computer interaction fields such as autonomous driving,virtual reality,and sign language translation.Gesture interaction technology mainly includes gesture recognition and hand pose estimation,which are interdependent and mutually reinforcing.At present,gesture interaction technology based on deep learning has excellent performance,but factors such as lighting,camera distance,and complex background have too strong interference on gesture data.At the same time,classical network models generally have a large number of parameters and computation,which affects the accuracy of gesture interaction and limits the application scenarios of the model.In response to the above issues,the specific research work of this article is as follows:(1)Aiming at the problem that the feature extraction ability of the network model is insufficient to reduce the accuracy of gesture recognition and increase the reasoning time of the model,a gesture recognition algorithm based on the improved MobileNet network is proposed.A multi-scale convolution module is designed to extract the underlying features,enhance the feature extraction ability of the network,and use the ELU activation function to retain more comprehensive negative feature information.The experimental results show that the accuracy of the proposed method on public datasets is higher than that of most lightweight network models,and the recognition speed can meet the real-time gesture interaction requirements;(2)A hand pose estimation algorithm based on HRNet is proposed to address the issues of increasing CPU computational power and unstable accuracy of hand pose estimation due to the large number of network model parameters.The algorithm adopts a parallel connection design of high and low resolution subnets to enhance hand pose feature representation,and combines the efficient channel attention module(ECA-Net)and the lightweight network module Geff of the Ghost module to replace the basic module in the original HRNet network.The experimental results show that the algorithm in this paper achieves the effect of relieving the pressure of equipment,reducing the amount of network parameters and improving the accuracy of key point estimation;(3)In order to improve the passive sign language gesture teaching mode,the gesture recognition network model is deployed on the Web side,and a WebAR assisted sign language gesture interaction system is built based on the MS-MobileNet network.The system has designed corresponding learning and usage modes for different user groups,achieving free conversion of text,speech,and gestures.After evaluation and testing,the practicality and effectiveness of the system have been verified,while exploring application scenarios for gesture interaction.