Traffic Sign Object Detection with the Fusion of SSD and FPN

To enhance traffic sign object detection further, this paper proposes a method that combines the SSD (Single Shot MultiBox Detector) model with FPN (Feature Pyramid Network) fusion for traffic sign identification. The SSD structure is leveraged to extract feature information from traffic sign images. Subsequently, the obtained feature information is fused using the FPN structure to enable more efficient object detection. Additionally, the attention mechanism is introduced to concentrate the fused FPN and SSD structures on target-relevant features, thereby improving accuracy, classification precision, and the ability to handle interference, ultimately enhancing overall performance. The post-processing step uses DIoU-NMS (Distance-IoU Non-Maximum Suppression) based on centroid distance. Validation is performed on the CCTSDB dataset, demonstrating a significant improvement in accuracy compared to traditional SSD. Experimental results confirm the effectiveness of the proposed method in enhancing detection efficiency and accuracy.