Project made by : Rehali & Kabouche

Project Overview

This project aims to create a reliable and comprehensive database for the YOLO model, containing traffic signs and traffic lights specific to Algeria. The YOLO model trained on this database can subsequently be integrated into an ADAS (Advanced Driver Assistance System) to detect vertical road signaling infrastructure using a camera.

Dataset

This project used the following databases as a fundamental basis, with modifications made to add or delete classes according to the desired final database :

For traffic signs :

Oprea, R., Traffic Signs Detection Europe Dataset, Roboflow, Editor. Feb 2024: Roboflow Universe.

For traffic lights :

Car, S., Self-Driving Cars Dataset, Roboflow, Editor. Jun 2023: Roboflow Universe.

Note : The traffic signs in this database originate from Romania, which has traffic signs similar to those found in France who are identical to those found in Algeria.

1. Description of the Classes Categories :

Our dataset contains 47 classes : 45 classes for traffic signs and 2 for traffic lights.

There are 4 main categories for the 45 traffic sign classes :

• Att --> (Attention/Warning)
• Inter --> (Interdiction/Forbidden)
• Oblig --> (Obligation/Mandatory)
• Indic --> (Indication/Information)

2. Dataset Optimization :

Ensuring the quality of the training database is crucial, as it directly impacts the results obtained during predictions with the trained model.

• Dataset Correction : We reviewed the entire dataset, which contained 4254 images before augmentation. Several errors were identified and corrected to improve the quality of the training data.

• Dataset Augmentation : To enhance the dataset, we added a significant number of images. These images were collected by us using our phones on Algerian roads, from Google image or through screens from YouTube videos depicting car trips in Algeria.Additionally, we used Google Street View with a screenshot extension to capture photos of traffic signs from various locations in France.