The Study Concerning Human-computer Interaction Used In The Manual Segmentation And Identification Of Key Techniques

With the rapid popularization of computer and Internet,human-computer interaction has gradually become an important part of people's daily life.It has been widely used in individual home,intelligent enterprise office,hospital security monitoring,intelligent education and other fields.It fills the gap between human and electronic devices.The traditional way of human-computer interaction,such as keyboard,mouse,remote control,touch screen,need to adapt to the machine,needs to operate,according to the preset specifications,and development of today's technology makes human-computer interaction have more choices.Based on hand gesture recognition,human-computer interaction mainly focuses on direct manipulation,which makes human-computer interaction technology transfer from machine centric to human centered.It tend to satisfy human communication habits.Therefore,gesture recognition founded on machine vision is being developed and applied to engineering field.This research includes gesture segmentation algorithm grounded on hybrid Gauss model and hand gesture classification algorithm derived on depth convolution neural network.This paper,at beginning,introduces the research background and significance of technology in the field of gesture recognition,and then prior research of status of gesture recognition algorithm is fully analyzed.Possible problems and improvement measures happened in the application process of gesture recognition algorithm are also described in details.The main contents of this paper include the following aspects:1.We introduce a problem in the process of gesture recognition,namely gesture segmentation,and segmentation method is proposed based on the mixed skin color model Gauss gestures,and eventually we had a full method of derivation and validation.2.We introduce the convolution neural network model for traditional image classification,and build up deep convolutional neural network used in end to end gesture classification.The model structure and parameters,activation function,cost function,training methods,test procedures and other details are also focused on in this paper.The selections of parameters in the process of network construction are given.3.Verification of the effect of gesture segmentation.Firstly,the introduction of a gesture database for validation of effective indoor and outdoor under two conditions is listed.Secondly,the gesture segmentation effect of indoor database is demonstrated,and then the average classification accuracy of the gesture before and after segmentation is compared.Then we compare the classification results of several classical convolutional neural network frameworks to illustrate the validity of our model,and the results and further classification of traditional features and traditional classifier algorithm based on the results of the comparison.Experimental results demonstrate the effectiveness of our approach.In the interior database,we will sign the correct classification rate is increased from 85.2%to 98.6%.Under outdoor conditions,National University Singapore public gesture database shown above,we will sign the correct average classification rate increases from 92.22%to 95.4%.Finally,we analyze the learning process and characteristics of deep convolution neural networks by visualizing the convolution kernels and feature maps of depth convolution neural networks.In this paper,we propose some solutions via the two key steps of hand gesture recognition,namely gesture segmentation and gesture classification,and verify the effectiveness of the algorithm through experiments.