Detection of People and Objects for the Visually Impaired by Using YOLOv7 Algorithm

This paper proposes a solution to an alarming social problem by proposing a system to assist the people with visual impairments. Blindness or vision impairment is among the top five disabilities of the world and has targeted nearly 18.7 million Indians. The absence of the ability to see the outer world and surroundings restricts a person’s mobility to a great extent and also creates possibilities of unfortunate accidents and harm to their life. Hence, there is an important need to develop a system that will guide, assist, and act as a virtual companion for the target category of people. In this paper, the focus is on building a prototype using computer vision algorithms that are composed of deep neural networks and perform object detection in live video streaming to assist an individual’s mobility in real world and surroundings by successfully detecting the objects and human figures and then converting the output to voice output that is fed into the target’s ear. The proposed work uses a fusion of the YOLOv7 object detection algorithm along with the PYTTSX3 conversion library that converts text into speech. YOLOv7 is probably the fastest object detection algorithm; thus, our contribution to the ongoing research is to present our ideas and optimization techniques for improving the accuracy and speed, reducing the cost, and introducing the concept of portability in blind assistance systems.