TSD-Net: A Traffic Sign Detection Network Addressing Insufficient Perception Resolution and Complex Background.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Intelligent Transportation Systems

Background:

• Detecting small traffic signs is vital for driving safety in intelligent transportation systems.
• Challenges include low resolution, dense distribution, and complex backgrounds, limiting current methods.
• Existing approaches suffer from high computational costs and feature alignment issues.

Purpose of the Study:

• To propose an improved framework, TSD-Net, for enhanced small traffic sign detection.
• To address limitations of existing methods in complex road environments.
• To improve accuracy and efficiency in detecting small traffic signs.

Main Methods:

• Developed TSD-Net integrating a Feature Enhancement Module (FEM) for expanded receptive fields.
• Introduced a high-resolution detection branch and Adaptive Dynamic Feature Fusion (ADFF) head for optimized feature fusion.
• Incorporated C3k2 module and dynamic convolution for flexible feature extraction and computational efficiency.

Main Results:

• TSD-Net demonstrated superior performance in small object detection and complex background handling on the TT100K dataset.
• Achieved 91.4 mAP and 49.7 FPS on low-resolution images (640 × 640).
• Outperformed most existing methods in critical detection tasks.

Conclusions:

• TSD-Net effectively enhances the detection of small traffic signs in complex environments.
• The framework meets practical application requirements for intelligent transportation systems.
• Offers a promising solution for improving road safety through advanced computer vision techniques.