Road Sign Detection Algorithm Based on Improved YOLOv8n

Nowadays, Traffic sign detection is gaining significance in the realm of smart driving such as autonomous driving and assisted driving. To improve the effectiveness of traffic sign detection algorithms in the area of traffic sign recognition, I propose an algorithm based on improved YOLOv8 to solve problems such as complex image backgrounds and small detection targets. Firstly, in order to reduce training time while maintaining performance, we proposed RepVitBlock on the basis of the lightweight structure Repvit, replacing the Bottleneck structure. Secondly, We have developed a lightweight PC2f module to substitute the C2f module in the original Neck section. We also introduced SIoU as a detection box loss function to improve the modal's performance. SIoU can better identify targets. Furthermore, we incorporated the SimAm attention mechanism to produce attention weights by assessing the local self-similarity of feature maps. Finally, we added a small object detection layer to extract shallow features on high-resolution images. After processing the traffic sign dataset CCTSDB, the results showed that the improved YOLOv8 algorithm achieved an accuracy of 82.4%, a recall rate of 74.1%, and an average accuracy of 86.7%. These results validate the effectiveness of our proposed model in traffic sign detection.