Automated Walking Guide to Enhance the Mobility of Visually Impaired People

This paper has shown the implementation detail of a spectacle prototype to assist the visually impaired people with safe and efficient walking in the surrounding's environment. The walking guide uses three pieces of ultrasonic sensors to identify the obstacle in three directions: front, left and right. In addition, the system can detect potholes on the road surface using another ultrasonic sensor and convolutional neural network (CNN). The CNN runs on an embedded controller to identify obstacles on the surface of the road. Images are trained initially using a CNN on a host computer and are then classified on the embedded controller in real-time. The experimental analysis reveals that the proposed system has 98.73% accuracy for the front sensor with an error rate of 1.26% when the obstacle is at 50 cm distance. In addition, the proposed system obtains the accuracy, precision and recall of 92.67%, 92.33% and 93% respectively for image classification. The experimental study also demonstrates how the developed device outperforms prominent existing works.