Scholarlink

<p>Vehicle detection under low-light conditions remains a significant challenge in the field of computer vision, exerting a notable impact on crucial applications such as autonomous driving and surveillance systems. Although the existing deep learning-based detection methods have achieved remarkable success under normal lighting conditions, their performance degrades significantly in low-light environments due to issues like insufficient brightness, low contrast, and loss of detailed features. This paper presents LLD-YOLO, an enhanced YOLOv11 for low-light vehicle detection. It incorporates improvements from a DarkNet module adapted from Self-Calibrating Illumination Learning for enhancing low-light images via adaptive illumination adjustment, a C3k2-RA feature extraction enhancement module that combines convolutional operations with self-attention mechanisms to overcome local receptive field limitations and capture global contextual information, and a Con-AM feature fusion module that optimizes multi-scale feature integration through an attention mechanism for adaptive feature selection and enhancement. Extensive experiments on Exdark demonstrate that our proposed LLD-YOLO achieves superior detection performance compared to existing methods, with significant improvements in detection accuracy and robustness under various low-light conditions. The mean average precision (mAP) of our method reaches 83.3%, which is a 4.5% improvement over the baseline model, while maintaining efficient computational performance.</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

LLD-YOLO: a multi-module network for robust vehicle detection in low-light conditions

  • 期刊论文

  • 发表日期：

    2025年2月4日

查看全文
  • Qin Zhang,
  • Weian Guo,
  • Meibin Lin

摘要

Vehicle detection under low-light conditions remains a significant challenge in the field of computer vision, exerting a notable impact on crucial applications such as autonomous driving and surveillance systems. Although the existing deep learning-based detection methods have achieved remarkable success under normal lighting conditions, their performance degrades significantly in low-light environments due to issues like insufficient brightness, low contrast, and loss of detailed features. This paper presents LLD-YOLO, an enhanced YOLOv11 for low-light vehicle detection. It incorporates improvements from a DarkNet module adapted from Self-Calibrating Illumination Learning for enhancing low-light images via adaptive illumination adjustment, a C3k2-RA feature extraction enhancement module that combines convolutional operations with self-attention mechanisms to overcome local receptive field limitations and capture global contextual information, and a Con-AM feature fusion module that optimizes multi-scale feature integration through an attention mechanism for adaptive feature selection and enhancement. Extensive experiments on Exdark demonstrate that our proposed LLD-YOLO achieves superior detection performance compared to existing methods, with significant improvements in detection accuracy and robustness under various low-light conditions. The mean average precision (mAP) of our method reaches 83.3%, which is a 4.5% improvement over the baseline model, while maintaining efficient computational performance.