Blind Navigation On Mobile Terminal Based On Deep Learning

With the rapid development of deep learning technology in recent years and the continuous update of mobile devices,great changes have been brought to people's lives,which has also made it possible for people with visual disabilities to travel safely and conveniently.The number of blind people in China has always been high,and the travel problem of blind people has attracted much attention.Researchers at home and abroad have developed a variety of scientific and technological products to assist the blind in traveling.However,these products generally have disadvantages such as single function,low cost performance and troublesome carrying,making them difficult to be used as long-term auxiliary traveling tools for the blind.In order to solve this problem,this paper designs a blind navigation system based on Android mobile devices.Blind users can realize voice navigation and obstacle avoidance functions only with an Android phone connected to the Internet,which not only ensures their travel safety,but also reduces the cost of use,and provides a new idea for blind navigation in the future.The main work of this paper includes:(1)Realize target detection function on mobile terminal.Most of the target detection network runs on the server,which is not convenient to carry and can not realize the real-time detection function of a single computer.In this paper,the model transformation method is adopted to make the target detection network run across platforms and realize real-time detection on the mobile end.This method reduces the cost of the obstacle avoidance function of the guide device and is easier to be accepted by the blind.(2)Improve the target detection network.In this paper,YOLOV4 network is chosen as the basic network.The huge computing amount of the basic network is suitable for running on the server side,but not on mobile devices with limited computing power.From the perspective of reducing the amount of computation,this paper improves the PANET part of YOLOV4 to reduce the amount of network computation.According to the test results,the improved network not only retains the identification accuracy of the original YOLOV4 network,but also improves the number of detected frames per second,so that the network can run more smoothly on mobile terminals.(3)Designed voice interaction and map navigation functions according to the travel needs of the blind.Blind people cannot operate mobile phones through their eyes,and common maps are not suitable for blind people.This paper uses iFlytek to realize voice interaction,and designs a walking navigation system suitable for the blind by calling Auto Navi SDK.