Development of perceptual aid for visually challenged persons using deep learning

Visual impairment is a serious problem that restricts a person from navigating, interacting with the surroundings, identifying objects and avoiding obstacles. There are several such challenges that a visually challenged person faces in day-to-day life and there are many technology-aided solutions currently available to assist them. But not all are highly reliable in different scenarios and affordable by everyone. In this paper, a low cost, compact, ergonomic deep learning driven perceptual aid for visually challenged persons with audio feedback is proposed. A camera attached to the device provides the live stream of the surrounding, to which a YOLO object detection algorithm is applied to detect the objects. An ultrasonic sensor attached near the camera provides the distance measurements of the objects within its detection range. From the distance measurements, the object that is nearest to the user is filtered and a text-to-speech conversion algorithm gives audio feedback to the visually challenged person via a Bluetooth-enable earpiece about the name of the object with its distance apart from the user wearing the aid. The proposed visual aid can be worn around the neck using a tag like an identity card and is efficient in detecting objects that are at a lower height. Compared to existing visual aids like generic white canes or electronic travel aids, the proposed visual aid has better accessibility, more comfort and makes the user more conscious about the surrounding.