Scholarlink

<p>Vehicle detection on expressways during twilight is challenged by low light, inoperative street lamps, and severe glare from different light sources. To address these challenges, we propose LLTD-YOLO, a vehicle detection method for expressways during the twilight period. The framework utilizes GAN-RetinexNet for image enhancement, incorporates iRMB attention mechanisms for robust feature extraction, and combines Focal Loss with bounding box regression to enhance detection accuracy in challenging lighting conditions. The experimental results demonstrate that LLTD-YOLO outperforms other models on our expressway dataset, with improvements of 0.4, 6.6, and 5.2% in precision, recall, and mAP@0.5, respectively. It also achieved the best performance among all the compared models on the ExDark dataset. In traffic flow detection, LLTD-YOLO reduced the average error during twilight to 2.5% (a 54.4% reduction) and across all periods to 2.3% (a 30.3% reduction), confirming its efficacy in twilight environment.</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

LLTD-YOLO: an image enhancement-based vehicle detection method for expressways in twilight environment

  • 期刊论文

  • 发表日期：

    2025年11月10日

查看全文
  • XiaoQi Yang,
  • JingYi Mao,
  • SiKai Kong,
  • PeiQun Lin

摘要

Vehicle detection on expressways during twilight is challenged by low light, inoperative street lamps, and severe glare from different light sources. To address these challenges, we propose LLTD-YOLO, a vehicle detection method for expressways during the twilight period. The framework utilizes GAN-RetinexNet for image enhancement, incorporates iRMB attention mechanisms for robust feature extraction, and combines Focal Loss with bounding box regression to enhance detection accuracy in challenging lighting conditions. The experimental results demonstrate that LLTD-YOLO outperforms other models on our expressway dataset, with improvements of 0.4, 6.6, and 5.2% in precision, recall, and mAP@0.5, respectively. It also achieved the best performance among all the compared models on the ExDark dataset. In traffic flow detection, LLTD-YOLO reduced the average error during twilight to 2.5% (a 54.4% reduction) and across all periods to 2.3% (a 30.3% reduction), confirming its efficacy in twilight environment.