AI / Data ScienceSHIPPED

AI for Helmet, Car, Motorcycle, and Number Plate Detection

Development of Artificial Intelligence for object detection using the YOLOv5 algorithm trained on 250 labeled images.

Timeline

2023

Role

AI / ML Engineer

Team

3 people

View Result

AI for Helmet, Car, Motorcycle, and Number Plate Detection – screenshot 1

OVERVIEW

A comparative study using YOLOv5m to detect helmets, cars, motorcycles, and number plates in images. Only 250 labeled images were available, so careful labeling and consistent evaluation were essential. The model was trained on Google Colab in Python, and results were evaluated with accuracy, precision, recall, F1-score, and mAP, then analyzed with error matrices, precision-recall curves, and F1 curves.

THE CHALLENGE

Only 250 images were available, so every label had to be accurate and consistent.
The model had to detect several object classes of different sizes (helmets, cars, motorcycles, and number plates) in one image.

THE SOLUTION

Labeled and classified all 250 images with Roboflow.
Trained a YOLOv5m model in Python on Google Colab.
Tested different numbers of training epochs and learning rates to see their effect on detection accuracy.
Evaluated results with accuracy, precision, recall, F1-score, and mAP, and analyzed them with error matrices, precision-recall curves, and F1 curves.