FA ‐ YOLO : An Enhanced YOLO Network for Pedestrian and Vehicle Detection

ABSTRACT Real‐time precise vehicle–pedestrian detection is critical for autonomous driving systems. However, existing methods often struggle to achieve an optimal trade‐off between detection accuracy and inference speed. To address this limitation, this paper proposes FA‐YOLO, an improved vehicle–pedestrian detection model based on YOLOv10n. First, the original partial self‐attention (PSA) module in the backbone network is replaced with a multi‐scale attention mechanism (MSAM) module, which enhances multi‐scale feature fusion while reducing computational redundancy and network parameters. Second, the spatial pyramid pooling fast average pooling (SPPFAP) module incorporates a global average pooling branch to effectively capture feature correlations and dependencies, thereby minimizing background interference. Additionally, a separated and enhancement attention module (SEAM) is integrated between the neck and head networks to emphasize target features and improve model robustness in occluded and complex background scenarios. Comprehensive experiments are conducted on the KITTI and BDD100K datasets to evaluate the proposed method against existing benchmark algorithms. The experimental results demonstrate that, compared with the original YOLOv10n, FA‐YOLO improves mAP0. 5 by 2. 5% and FPS by 6. 1% on the KITTI dataset, and achieves respective improvements of 1. 2% and 1. 7% on the BDD100K dataset. Furthermore, when compared with other state‐of‐the‐art approaches, FA‐YOLO offers substantial advantages in both detection accuracy and computational efficiency, making it particularly well suited for target detection tasks in autonomous driving scenarios.