THE DETECTION METHOD OF SMALL TARGET TRAFFIC SIGNS IN COMPLEX SCENARIOS

For the problem of small-target traffic signs being easily missed and mis-detected in complex scenes, a small-target traffic sign detection method based on the improved YOLOx-s model is proposed. The improved YOLOx-s model used ConvNext-T as the backbone feature extraction network and combined Meta-ACON, RFP, and Focal-EioU methods for end-to-end training of the data-enhanced CCTSDB dataset. It was achieved to improve the detection performance of the model for small targets without increasing the attention mechanism, while maintaining the simplicity of the model. The experimental results show that the improved model pays more attention to small traffic signs in the target sample, and the mAP value improved by 5.31 percentage points.