Assistive Eye: A Comparative Analysis of YOLO Object Detection Models on Edge Devices

This paper aims to provide an in-depth analysis of the performance of various versions of the YOLO (You Only Look Once) object detection models and aims to provide a comparative perspective on the functioning of these models on a Raspberry Pi 8GB RAM version. As a segment of a broader research project aimed at developing the Assistive Eye system for aiding visually impaired individuals in navigation, this study narrows its focus to the initial observations and comparative performance analysis of these models. We explore the potential of running different iterations of the YOLO models including the YOLOv8n (nano), the latest low-compute demanding model. In this paper, we also go through the challenges, effectiveness, memory usage, and computational intensity of each model in inferring objects in real-time. The eventual goal is to integrate the most suitable model with a text-to-speech system for auditory feedback. In the scope of this paper, we will solely concentrate on the object detection capabilities and performance metrics of the YOLO models in this initial phase of research.