NETRA: A Revolutionary Navigation Aid for the Visually Impaired

In India, approximately 0.36% of the total population is visually impaired or blind. These visually impaired people constantly need a guardian to help them, and use of a normal white cane restricts their mobility in unfamiliar environments. Even though there are smart canes available in the market, they are not trained to work in Indian conditions. In this research paper, we explore the use of object detection, image segmentation, and captioning in order to generate reliable audio instructions, which will help the blind person in navigation. This paper proposes a design for a smart cane that includes image sensors, ultrasonic sensors, and SOS button mounted on it, which will be working along with the smartphone camera in order to provide the live feed video of the surroundings the user is in. The YOLOv7 model, which works on the custom dataset built on the COCO dataset, helps in object detection. The paper emphasizes the way we can use deep learning algorithms to solve the problem of difficulty in mobility of the visually impaired.