Attention feature fusion network for small traffic sign detection

Abstract Object detection has made great progress with the rise of convolutional neural networks in recent years. Traffic sign detection is a research hotspot for object detection tasks. The existing detection models have the problems of inaccurate positioning and low classification accuracy when detecting small traffic signs. To address these issues, in this paper, we propose a small traffic sign detection method based on YOLOv4. Specifically, we design an attention-based feature fusion module including attention spatial pyramid pooling (ASPP) and attention path aggregation networks (APAN). ASPP highlights useful small object information and suppresses invalid interference information in the background. APAN reduces information loss during feature fusion. A large number of experimental results on public datasets show that the method in this paper improves the detection performance of the model. In terms of small traffic sign detection, the method improves YOLOv4 by 12 mAP, and meets the real-time requirements of automatic driving detection (more than 50 FPS).