Research On Gesture Recognition Algorithm Based On Convolutional Neural Network

With the continuous development of computer science and the rapid progress of internet technology,gesture recognition technology has been widely used in human-computer interaction,service industry,information industry and other fields,and has gradually become a research hotspot.Therefore,the research on gesture recognition algorithm has important theoretical research significance and application value.However,the traditional gesture recognition method has the disadvantages of low recognition rate and long time consumption.This paper uses the advantages of convolutional neural network to automatically extract and learn image features,and combines foreground detection and migration learning algorithms to design a convolutional neural,The gesture recognition method of the network improves the recognition rate and training efficiency.First of all,in terms of moving target foreground detection algorithm,in the traditional background difference method,when detecting the foreground of the gesture,the contour extraction is unclear,which affects the accuracy of gesture recognition,etc,this paper improves the background reconstruction and threshold selection methods,by comparing experiments It shows that the improved background difference method improves the accuracy rate of gesture recognition by 2.22 percentage points.In terms of selecting the network structure,the traditional Lenet-5 network structure is relatively simple.When processing data sets with complex texture features,there is a problem that the classification accuracy is not high.In this paper,the network structure is improved,using the idea of cross-connection and increasing the number of convolution kernels and fully connected layer neurons,so that the network can extract more features.Compared with the traditional Lenet-5 network structure,the recognition accuracy has been improved by nearly 6%.Then,for the problems of deep learning hardware resource requirements and long training time,this paper introduces a transfer learning algorithm to shorten the training time by nearly half.In view of the influence of network model parameters on the final recognition rate,this paper compares different activation functions,optimizationalgorithms and Dropout coefficients respectively,and selects the most suitable model parameters for the experiments in this paper,which improves the recognition accuracy of the model.Experimental results show that the gesture recognition algorithm based on convolutional neural network proposed in this paper has greatly improved the recognition rate and training efficiency compared with other methods.