In the era of big data, vehicle detection has been a highly complex and widely discussed topic. With the acceleration of urbanization and the increase in traffic volume, the management and monitoring of road traffic have become increasingly urgent, highlighting the growing importance of vehicle target detection. However, existing algorithms suffer from issues such as low accuracy, high false negative rates, and poor robustness. Therefore, this study introduces an improved vehicle detection algorithm, GPF-YOLO, based on the YOLOv8 model. This algorithm integrates the GD mechanism, polarized self-attention (PSA), and the optimized loss function Focaler-IoU, aiming to enhance target detection accuracy. The GD mechanism improves multi-scale feature fusion capabilities, enhancing the detection of targets at different scales. Polarized self-attention introduces a Polarized Self-Attention mechanism that effectively handles fine-grained information about target locations, especially in pixel-level regression tasks. Focaler-IoU optimizes the bounding box regression loss function, focusing on the impact of challenging and straightforward sample distributions on regression results, further improving target detection performance. Experimental results demonstrate that GPF-YOLO outperforms the YOLOv8 model with a 2.05% improvement in mAP@50. Our approach surpasses other target detection algorithms, significantly enhancing the overall system’s robustness and accuracy. It provides a higher level of precision in vehicle detection, offering critical technical support and references for real-time target detection tasks.