Android Platform Oriented Deep Learning Based Key Technologies Study And Implementation For A Blind Guide System

The vast majority of blind people lack guidance without being accompanied,making it difficult to go out alone.As an important infrastructure of the city,blind roads have not played their due role.Therefore,how to effectively use blind information to help blind people travel has become the focus of scholars.As the society pays more and more attention to the visually impaired people,the research and development on blind guide systems becomes very practical.This dissertation discusses and studies the two key technologies of blind guide system blind path segmentation and obstacle detection,and the main research content and results are listed as follows:1.A blind tracks segmentation model based on improved U-net for Android system is constructed.This method is improved on the basis of U-net network by reducing compression blocks,expansion blocks,and the number of channels of the model convolution layer,to build a blind tracks segmentation model suitable for running on mobile devices.The experimental results show that the constructed model has good blind tracks segmentation results on PC and mobile terminal,and the segmentation speed on the mobile terminal is increased by more than 10 times compared with the original U-net.2.An Obstacle detection model based on convolutional neural network for Android system is constructed.Considering the poor effectiveness of traditional methods for detecting obstacles and the limited resources of mobile devices,this dissertation designs an obstacle detection model with few parameters and high detection performance.The experimental results show the effectiveness of the model.The AUC value is 92.5%,which is 10% higher than the method in [10],and the detection speed on the mobile terminal is much faster than the method in [10].3.Transplant the trained blind tracks segmentation model and obstacle detection model to the Android mobile operating system platform,and develop a blind guide APP.We test the running speed and results of the APP on mobile phone devices.Tests show that the APP can be used on mobile phone devices effetively,accurate blind tracks segmentation,and good obstacle detection effect.