Comparison of YOLO Models for Object Detection from Parking Spot Images

Abstract

This paper compares several You Only Look Once (YOLO) models for object detection in parking lot images. Surveillance, independent vehicles, and intelligent cities are some of the applications that demand object recognition. The YOLO algorithm has undergone various iterations to improve real-time performance and increase accuracy. Effective parking space management is one of the major players in reducing traffic congestion in cities. Computer vision-based systems show us a way forward by automatically identifying free parking slots. The present paper compares using the YOLOv3, YOLOv5, YOLOv7, and YOLOv8 models to test images taken from car parks in different periods of the year and under various meteorological conditions. The results of the conducted tests show all the strengths and weaknesses of each YOLO model type, considering statistical elements such as precision, recall, evaluation time, and ease of use.