CT-YOLO: improved YOLOv8 for object detection on traffic signs

As the demand for intelligent transportation systems and autonomous driving technology increases, accurate traffic sign detection becomes a critical component for road safety and autonomous navigation. However, challenges such as class imbalance, varying sign shapes, and image distortions significantly affect detection performance in real-world scenarios. To address these issues, this study proposes an enhanced YOLOv8-based model for traffic sign detection, incorporating novel techniques to improve accuracy and robustness. Firstly, a data augmentation method is introduced that generates additional training data for underrepresented sign categories by overlaying traffic sign templates onto background images, effectively tackling class imbalance. Secondly, a Color-Position Attention (CPA) mechanism is employed to improve the model's ability to focus on significant color and spatial features, enhancing object localization. Finally, Deformable Convolutions (DCN) are integrated to address the challenges posed by non-rigid and distorted traffic signs. The proposed model significantly outperforms existing object detection models achieving superior performance in terms of accuracy, recall, and mean Average Precision (mAP), particularly in complex, real-world detection scenarios.