Dynamic Gesture Recognition Method Based On Convolutional Neural Network

With the continuous development of emerging information technologies such as artificial intelligence,cloud computing,and the Internet of Things,human-computer interaction using gestures has become a research hotspot in the field of computer applications.Establishment of a good human-computer interaction environment is becoming the important technical needs and trends of the next-generation wireless that supports the Internet of Everything.However,with the complexity of the scenes used for gestures,gesture recognition facing the camera at close range can no longer meet human needs.Factors such as distance and angle have a greater impact on recognition.When gestures are rotated or even overlapped between fingers in the collected pictures,the static gesture recognition method cannot achieve good results.This paper studies the problem of low accuracy of gesture recognition under the condition of rotating occlusion,and proposes a dynamic gesture recognition method based on convolutional neural network:(1)The principle of the inter-frame difference method is introduced,and the idea of using the inter-frame difference method to extract key frames from a video sequence is explained.First the differences between every two frames of the video sequence are calculated.Then the improved Hanning window filtering is used to smooth the scatter plot and the frame of the image maximum is extracted as the key frame.Then YCb'Cr' color space is used to perform gesture extraction on the key frame image.Finally,the key frame image after the gesture is extracted is stitched with the final gesture image as gesture data.Experimental simulations verify that the method can obtain effective gesture regions.(2)Aiming at the limitation of traditional data enhancement methods,a data enhancement method based on Variational Auto-Encoder is proposed.The common data enhancement methods such as geometric transformation and noise addition are introduced,and their respective limitations are analyzed.To solve this problem,Variational Auto-Encoder is used to learn and encode the original data and generate new data to achieve the effect of data enhancement.(3)A convolutional neural network network framework for gesture recognition and classification is built.In response to the problem of data image format,the convolution method of the first convolution layer was improved.Finally,the effectiveness of the new method is verified by experiments.At the same time,different data selection methods and different convolution methods are compared and verified,and the superiority of the method is verified.