Faster-YOLOv8n-based lightweight road sign detection for intelligent unmanned vehicles

Existing lightweight road sign detection methods can hardly address the challenges of occlusion and scale difference. Therefore, a Faster-YOLOv8n based lightweight road sign detection method is proposed in this paper for intelligent unmanned vehicles. Firstly, we introduce Partial Convolution (PConv) into traditional YOLOv8n. The bottleneck of the C2F module is replaced by FasterBlock, which reduces the number of parameters in the model and improves the speed. To reduce the accuracy loss due to lightweighting and scale differences in the near and far views, a Multi-scale Feature Interaction Attention (MSFIA) is proposed in the feature fusion neck network. In addition, a Spatially Reinforcement Attention (SRA) module is proposed at the detection heads of the network. SRA strengthens the response to unoccluded regions and weakens the background regions to address the occlusion problem. We conducted comprehensive comparative experiments on the TT100K dataset. The experimental results show that compared to the original YOLOv8n, the proposed Faster-YOLOv8n has a substantial improvement in both detection accuracy and inference speed.