Abstract: To realize road target detection in scenes with dense targets and complex backgrounds during autonomous driving, an improved YOLOv7 tiny detection algorithm is proposed. First, the SiLU activation function was used to replace the Leaky ReLU in the original network to enhance the feature extraction ability of the model. Second, a small object detection layer was added to improve the detection accuracy, and multi-scale information was captured by introducing a Receptive Field Enhancement (RFE) module. Finally, the backbone network was improved to a Dense Channel Compression for Feature Spatial Solidification Structure (DCFS) to improve the purity of the forward propagation features. Simultaneously, the gradient flow in the backpropagation was enhanced. By combining the advantages of different imaging modalities, experiments were conducted on infrared and visible image fusion datasets. The results showed that the F1 score of the improved algorithm was 80.6, mAP@50 was 84%, and mAP@50-95 was 51.2%, that were increased by 3%, 4.6%, and 3.6%, respectively. The improved algorithm could effectively improve the accuracy of road target detection and solve the missed and false detection problems.