Scholarlink

<p>In the field of traffic, intense sunlight presents a significant hazard to driving safety as it impairs drivers’ vision and increases the likelihood of traffic accidents. The current datasets lack images of traffic signs under glare conditions, highlighting the urgent need for a dedicated glare traffic sign dataset. To address this issue, we have developed the Glare Traffic Sign Dataset (GTSD), which includes comprehensive annotations of solar glare and encompasses various glare conditions encountered in everyday driving. Additionally, we propose the YOLO-SFM model for detecting traffic signs under glare conditions. YOLO-SFM utilizes SPD-Conv downsampling to minimize information loss and enhance feature extraction accuracy. The model integrates MLCA and FasterNet modules to merge local and global features, as well as channel and spatial information, thereby reducing redundant computations and memory access while improving detection accuracy and speed. Compared to existing mainstream object detection models, YOLO-SFM demonstrates superior detection performance on both the GTSD and CCTSDB2021 datasets. On the GTSD dataset, YOLO-SFM improves mAP at 0.5 intersection over union (IoU) threshold (mAP@50) by 11.8% compared to the baseline model while also reducing parameters by 0.3M.</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

Traffic sign detection in glare environments

  • 期刊论文

  • 发表日期：

    2025年6月10日

查看全文
  • Jie Bai,
  • Fengping Wang,
  • Haoqi Wang,
  • Jinhao Guo

摘要

In the field of traffic, intense sunlight presents a significant hazard to driving safety as it impairs drivers’ vision and increases the likelihood of traffic accidents. The current datasets lack images of traffic signs under glare conditions, highlighting the urgent need for a dedicated glare traffic sign dataset. To address this issue, we have developed the Glare Traffic Sign Dataset (GTSD), which includes comprehensive annotations of solar glare and encompasses various glare conditions encountered in everyday driving. Additionally, we propose the YOLO-SFM model for detecting traffic signs under glare conditions. YOLO-SFM utilizes SPD-Conv downsampling to minimize information loss and enhance feature extraction accuracy. The model integrates MLCA and FasterNet modules to merge local and global features, as well as channel and spatial information, thereby reducing redundant computations and memory access while improving detection accuracy and speed. Compared to existing mainstream object detection models, YOLO-SFM demonstrates superior detection performance on both the GTSD and CCTSDB2021 datasets. On the GTSD dataset, YOLO-SFM improves mAP at 0.5 intersection over union (IoU) threshold (mAP@50) by 11.8% compared to the baseline model while also reducing parameters by 0.3M.