Research On Dynamic Gesture Recognition And Gesture Estimation Methods Based On Convolutional Neural Network

With the rapid development of artificial intelligence,gesture recognition and gesture estimation are becoming more and more important in the field of computer vision.For example,people are no longer satisfied with a single human-computer interaction,and gesture tasks provide a new way for human-computer interaction.However,whether the effects of gesture recognition or gesture estimation can not really meet the needs of a real gesture system,so how to better complete the gesture task is still a problem that cannot be ignored in the field of computer vision.With the rapid development of deep neural networks,gesture-related technologies have become more mature.Therefore,this paper combined with the knowledge of deep learning and computer vision to better complete the task of gesture recognition and gesture estimation.Specific research contents are as follows:Firstly,around gesture recognition and gesture estimation,we understand the related concepts,and summarize the basic concepts and related knowledge.At the same time,the related concepts of convolution neural network and the structure of convolution neural network involved in this topic are briefly introduced.Secondly,a dynamic gesture recognition method based on convolutional neural network is proposed.The multi-scale empty convolution structure is used to increase the sensing field of the network and extract the information of different scales,so as to obtain the feature map with more effective information,which is then input into the compressed RPN and RCNN joint network for further feature extraction,and the static gesture recognition model is obtained and the static gesture recognition is carried out.Finally,the dynamic gesture recognition algorithm is designed based on the characteristics of time series,gesture category,gesture coordinates and gesture area.Thirdly,a three-dimensional gesture estimation method based on convolutional neural network is proposed.Firstly,the spatial dimension of the gesture segmentation network is increased by using the feature fusion of the deforming convolution to improve the segmentation effect of irregular gestures.Then the SENet-Inceptionv3 structure is used to enhance the attention between the channels of the feature map and obtain the heat map of 2d key points.Finally,the final 3d gesture key point estimation results are obtained by using the prior knowledge from the 3d gesture estimation network regression.Finally,the proposed method of gesture recognition and estimation is implemented and the corresponding experiments are carried out,and the proposed method is compared with the experimental results of the existing methods.