Research On Tactile Vision Substitution Technology Based On Convolutional Neural Network

According to the statistics of China Disabled Persons Federation,there are approximately 13 million visually impaired patients in China.The number of blind people is about 5.5 million.It is estimated that the number of newly added blind people will be around 450,000 every year.Although blind people can't see,the blind people's tactile system is very sensitive.The tactile vision substitution technology takes full advantage of blind people's sensitive haptic systems.The tactile vision substitution technology delivers visual information to the blind through haptics,which enables blind people to “see” this world in a tactile manner.The traditional tactile vision substitution technology has the disadvantages of poor tactile perception,low image recognition accuracy,and limited tactile resolution.In order to overcome the shortcomings of traditional tactile vision substitution technology,this paper uses Convolution Neural Network technology to improve the traditional tactile vision substitution technology,which greatly improves the effect of blind people's tactile perception.The main research content of this article is as follows:1、This paper studies the edge detection technology of visual images.The visual image contains a large amount of image information,which is not conducive to blind people's tactile perception.The visual image must be simplified,discarding the redundant information of the image.In this paper,the edge detection method is used to simplify the visual image and convert the visual image into the edge image.The simulation results show that the edge image can effectively reduce the amount of information of the visual image.The edge image only retains the most critical edge information of the visual image.Therefore,the edge image is more conducive to blind people's tactile perception.2、This paper studies the recognition technology of visual images.It is difficult for a blind person to determine the category of an image by sensing the edge image.The visual image must be classified and recognized,so that the blind person can directly perceive the category of the image through touch.This paper uses Convolution Neural Network technology to identify visual images.This article improves the traditional LeNet-5 model.This paper improves the network's activation function,pooling method,number of convolution kernels,output layer structure,and layer-to-layer connection.Finally,the improved LeNet-5 model achieves higher accuracy of digital recognition.The training method of the improved LeNet-5 model is studied.The experimental results show that using the improved LeNet-5 model can make the blind person accurately identify the current figures,and using the GoogLeNet model can make the blind person accurately identify everyday objects in front of them.3、This article designed a software platform for image acquisition and processing.The software platform uses a CCD camera to capture image information.The software platform calls OpenCV related libraries for edge detection of visual images.The software platform calls the improved LeNet-5 model and GoogLe Net model for image recognition of visual images.4、This article designs a device for generating electrical haptics.The tactile stimulation used in this paper is electrical stimulation.The image acquisition and processing section sends the processing result of the visual image to the electric haptic device.The electric haptic device maps the edge image and the recognition result of the image to the electrode array.Therefore,the blind person can not only perceive the edge information of the image but also recognize the category of the image.