Development Of Blind Lane Detection Device Based On Image Processing

China was the country with the largest number of blind people in the world.Due to the loss of visual perception,it was difficult for blind people to obtain all kinds of information in the environment.And they faced great difficulties in daily travel.Although the existing guide tools provide some help for the blind to travel,there are some problems of these tools in the process of actual use,such as expensive,single function,poor realtime performance.Based on this,this paper studied and designed a blind way detection device with blind guide function.In order to help the blind to walk on the blind way safely and improve the utilization rate of the blind way,the functions required for design of the blind way detection device include blind way recognition,safety obstacle avoidance and voice prompt.The blind way recognition function was realized by semantic segmentation technology based on depth learning Deep Lab V3 + to realize blind way image segmentation.Then,it extracted the feature information of two boundary lines of the blind way from the segmented blind way image by Gaussian filter,Canny edge detection,and Hough transform.Finally,it calculated the angle from which the blind person was deviated from the blind way based on the characteristic information extracted,and provided the corresponding voice prompts.The safety obstacle avoidance function was realized by ultrasonic sensors to collect obstacle data,transit time measurement method to calculate the distance information of obstacles,and voice prompts blind people to avoid obstacles.The blind way detection device designed in this paper used a small embedded development board Jetson nano as the main control unit and Open CV,Python,and Pytorch as the system development environment.First,the blind way data set was made and the network model was trained on PC.Then,the algorithm was transplanted to the operating system of the development board.The interface function in Open CV visual library was used to load the model and process-related image algorithms.The function and system performance of each module for blind way detection device was tested at last.The test results showed that the accuracy for blind way identification of the blind way detection device was more than 91%,and the processing time of a single frame is less than 0.25 seconds,which met the expected design requirements.