Traffic Sign Recognition for Autonomous Driving Using Branched YOLOv2 and Geometric Features

TL;DR

This paper proposes a traffic sign recognition system using a branched YOLOv2 architecture and geometric features. The system achieves a mean average precision of 0.680 and a runtime of 0.647 seconds. Geometric verification during inference increases mean average precision to 0.713.

Why It Matters

Accurate traffic sign recognition is crucial for autonomous driving and advanced driver-assistance systems. The proposed system can improve the safety and efficiency of these systems. The use of geometric features can reduce classification errors between visually similar signs.

Key Findings

• Branched YOLOv2 architecture reduces computation time
• Geometric features improve classification accuracy
• Unsupervised Bayesian image-segmentation method produces effective binary representations