Abstract: To address the limitations of target detection methods in complex traffic scenarios, particularly missed detections of small and occluded targets and challenges in multi-scale target detection and model robustness, we propose an improved YOLOv8s-SRCEM model. We introduce a small target detection head to enhance the sensitivity of the model to small-sized targets, improving its detection capability for such objects. Additionally, we integrate the Res-CBAM attention module into the small target detection head to further enhance feature learning salience. We incorporate the ECA module into the backbone network to strengthen the attention to important feature channels of the model and improve feature selection and model robustness. Furthermore, by replacing the original SPPF module with MS-Block, we enhance the feature capture and fusion capabilities of the model across different scales. On the KITTI dataset, the improved model achieves a 6.6% increase in mAP compared to the YOLOv8s model. Experimental results demonstrate that the combination of these enhancements substantially improves the detection performance of the model in complex traffic scenarios.