Design and Development of a Virtual Eye for Visually Impaired People in Indoor and Outdoor Applications

Abstract

Despite recent progress, people who are blind or visually impaired still face navigation difficulties, even in familiar surroundings. The current state-of-the-art approach to this problem is multi-sensor assistive technology, but additional sensors increase hardware cost and system complexity. This work proposes a camera-based navigation system that relies on vision alone and requires no additional sensors. The processing unit is the NVIDIA Jetson Nano, which can run real-time computer vision and deep learning algorithms. A single camera continuously captures image frames from the surrounding environment, and the processing module applies You Only Look Once (YOLO) object detection to the frames to recognize persons and objects such as vehicles, walls, furniture, and other obstacles. Audio instructions are delivered to the user's earphones as warnings or walking guidance. Qualitative tests showed that the system detected common objects in real time both indoors and outdoors, although detection performance dropped in dark conditions; detection accuracy and latency were not measured quantitatively. The single-camera design simplifies the system while still allowing users to understand their position relative to their surroundings.