Improved Traffic Sign Detection with ECA Attention and Augmentation

Traffic sign detection plays a vital role in Intelligent Transportation Systems (ITS) and Unmanned Vehicles (UVs), as it enables machines to perceive and interpret the road environment for safe and autonomous navigation. Accurate classification of traffic signs enhances driving safety, eliminates human dependence, and strengthens the decisionmaking capability of Advanced Driver Assistance Systems (ADAS). While traditional deep learning models such as YOLOv3 and YOLOv7 have shown strong performance in detecting and localizing road signs, they are often affected by challenges like lighting variations, motion blur, and adverse weather conditions. Moreover, existing attention mechanisms like Squeeze-and-Excitation (SE) introduce additional computational overhead. To address these limitations, the proposed study introduces an improved Traffic Sign Detection model based on YOLOv8, integrated with Efficient Channel Attention (ECA) and advanced data augmentation techniques. The system enhances generalization through synthetic weather degradation, varied intensity, and motion blur effects. Replacing SE with ECA significantly reduces inference time and model size without compromising accuracy. Trained with the AdamW optimizer and a Cosine learning rate scheduler, the model achieves an impressive mAP of about 99% at 0.5 IoU on the German Traffic Sign Detection Benchmark (GTSDB), surpassing baseline results in both accuracy and speed. Its lightweight and robust architecture ensures scalability, efficiency, and real-time performance, making it ideal for embedded automotive applications.