CW-DETR: An Efficient Detection Transformer for Traffic Signs in Complex Weather.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Intelligent Transportation Systems

Background:

• Adverse weather conditions like rain, fog, and snow degrade traffic sign features, challenging existing detection systems.
• Real-time traffic sign detection is crucial for intelligent transportation systems (ITS).

Purpose of the Study:

• To develop an end-to-end detection framework, CW-DETR (Complex Weather Detection Transformer), for robust traffic sign detection under adverse weather.
• To address weather-induced feature deterioration and improve real-time detection performance.

Main Methods:

• The CW-DETR framework builds upon RT-DETR, integrating four key innovations: multipath feature enhancement network (FPFENet), Multiscale Edge Enhancement Module (MEEM), adaptive dual-stream bidirectional feature pyramid network (ADBF-FPN), and multiscale convolutional gating module (MCGM).
• These modules are designed to preserve textures, enhance boundaries, compensate for cross-scale features, and reduce semantic-spatial confusion.

Main Results:

• CW-DETR achieved 69.0% AP and 94.4% AP50 on the CCTSDB2021 dataset, outperforming state-of-the-art real-time detectors by 2.3-5.7 percentage points.
• The model demonstrated computational efficiency with 56.8 GFLOPs.
• Cross-dataset evaluations on TT100K, TSRD, CNTSSS, and LNTU-TSD confirmed the model's robust generalization capabilities.

Conclusions:

• CW-DETR offers an effective solution for all-weather traffic sign detection.
• The proposed framework significantly enhances detection performance in challenging weather conditions.
• CW-DETR demonstrates strong generalization, making it suitable for real-world intelligent transportation applications.