Traffic sign classification using CNN and detection using faster-RCNN and YOLOV4

Area of Science:

• Computer Vision
• Machine Learning
• Autonomous Systems

Background:

• The increasing prevalence of autonomous vehicles necessitates robust traffic sign recognition for enhanced safety.
• Accurate and fast identification of traffic signs is crucial for reliable autonomous driving systems.

Purpose of the Study:

• To develop and evaluate a Convolutional Neural Network (CNN) for classifying 43 types of traffic signs.
• To implement and compare Faster R-CNN and YOLOv4 for real-time traffic sign detection.
• To assess the generalization ability of the developed CNN under severe conditions.

Main Methods:

• A lightweight CNN was designed and trained on the German Traffic Sign Recognition benchmark dataset (39,186 training, 12,630 testing images).
• Faster R-CNN and YOLOv4 networks were utilized for traffic sign detection using the German Traffic Sign Detection benchmark dataset.
• Performance was evaluated using accuracy for classification and mean average precision (mAP) with Intersect Over Union (IoU) of 50% for detection, alongside Frames Per Second (FPS).

Main Results:

• The developed CNN achieved a high classification accuracy of 99.20% with only 0.8 million parameters.
• Faster R-CNN demonstrated a mAP of 43.26% at 6 FPS, deemed unsuitable for real-time applications.
• YOLOv4 significantly outperformed Faster R-CNN, achieving a mAP of 59.88% at 35 FPS, making it suitable for real-time detection.

Conclusions:

• The lightweight CNN is highly effective for traffic sign classification, demonstrating strong generalization capabilities.
• YOLOv4 is the recommended model for real-time traffic sign detection systems due to its superior performance in terms of mAP and FPS.
• Comparative analysis highlights the trade-offs between different deep learning models for traffic sign recognition tasks.