Real-Time Computer Vision Based Autonomous Navigation System for Assisting Visually Impaired People using Machine Learning

Visual impairment is a global problem and people without vision suffer more than other impaired people. A companion is always needed for the movement of blind people and there may not be anyone by their side in case of emergency. Walking alone on the street, detecting the person closest to him and avoiding obstacles are always problems. Researchers have been working for visually impaired people using sensor based distance measurement systems for years. This paper proposes a Computer Vision based system to navigate visually impaired people by using Artificial Intelligence, and also a novel distance measuring approach. The system will capture real time images through a camera placed inside a sun-glass then process the video frames by trained YOLO V3 model. After processing, the program will identify a total of 80 pre-trained objects and additional 7 objects including person, car, bicycle, and broken roads and then will produce a navigation command through headphones. A comparative evaluation with other similar works is performed, and the result represents the primary accomplishments of this article. Several testing and validation procedures were carried out in order to achieve optimal performance and accurate distance measurement. The proposed system outperforms the state-ofthe-art in terms of object detection, distance measurement, computational costs calculation, and accessibility for the visually impaired, according to the results, which were validated using mathematical calculations and the necessary measuring devices. Since Industry 4.0 demands smart automation, this system has a significant impact not just on disabled persons but also on the development of a smart city.