Traffic Sign Detection and Recognition System Using Yolov5 and Its Versions

Date of Award

12-1-2022

Document Type

Thesis

Degree Name

Master of Science in Computer Science

First Advisor

Patricia Angela R. Abu, PhD

Abstract

A Traffic Sign Detection and Recognition (TSDR) System, which helps navigate vehicles through computer vision, has to perform quickly as vehi- cles using them travel at high speeds. A speedy one-stage detector such as YOLOv5, a deep learning model, was chosen to dive into during this study. This study explores creating a one-stage TSDR model by comparing four different versions of YOLOv5, namely YOLOv5 Nano, Small, Medium, and Large. This study was accomplished by first creating a dataset and split- ting it into training, validation, and testing sets to be used in the study. The four versions of the YOLOv5 algorithm were then trained with a 75- 25 train test split, and 24 models were created. After creating the models, they were each tested on the test set, and their respective metrics were tal- lied. Results showed that YOLOv5 Medium and YOLOv5 Large offer a 10% increase in accuracy performance when compared to YOLOv5 Small, but due to the instability of the validation loss of YOLOv5 Large, the YOLOv5 Medium models appear to be a better fit when it comes to the detection of traffic signs when prepared by a relatively small dataset done during real- time testing.

Link to Full Text

Share

COinS