Scholarlink

<p>Real-time detection of small and occluded objects in complex road scenes remains highly challenging, as models must simultaneously achieve accurate recognition, high throughput, and low latency. To tackle this issue, we propose DFPD-YOLO, an enhanced YOLOv8-based detector optimized for deployment efficiency. The model integrates a feature-focused diffusion pyramid neck to reinforce multi-scale contextual representation, a dynamic task alignment detection head to leverage joint features with reduced computational cost, and the WIoU loss to enhance localization precision. Extensive experiments demonstrate that on the KITTI dataset, DFPD-YOLO outperforms the baseline YOLOv8n by 4.5% mAP@50 while sustaining 242.4 FPS, and on the BDD100K dataset, it achieves a 2.2% mAP@50 improvement with 295.7 FPS. These results confirm the model’s capability to deliver both high detection accuracy and real-time performance, making it well suited for deployment in intelligent transportation and road safety systems on high-performance platforms, highlighting its suitability for HPC-oriented real-time applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced road object detection with DFPD-YOLO: focusing on small and occluded targets

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Zifeng Liu,
  • Lujiao Li,
  • Yongbin Hu,
  • Shigang Hu

摘要

Real-time detection of small and occluded objects in complex road scenes remains highly challenging, as models must simultaneously achieve accurate recognition, high throughput, and low latency. To tackle this issue, we propose DFPD-YOLO, an enhanced YOLOv8-based detector optimized for deployment efficiency. The model integrates a feature-focused diffusion pyramid neck to reinforce multi-scale contextual representation, a dynamic task alignment detection head to leverage joint features with reduced computational cost, and the WIoU loss to enhance localization precision. Extensive experiments demonstrate that on the KITTI dataset, DFPD-YOLO outperforms the baseline YOLOv8n by 4.5% mAP@50 while sustaining 242.4 FPS, and on the BDD100K dataset, it achieves a 2.2% mAP@50 improvement with 295.7 FPS. These results confirm the model’s capability to deliver both high detection accuracy and real-time performance, making it well suited for deployment in intelligent transportation and road safety systems on high-performance platforms, highlighting its suitability for HPC-oriented real-time applications.