Scholarlink

<p>Independent walking poses a significant challenge for visually impaired individuals, especially during street crossings. The absence of visual cues like pedestrian lights and directional signs leads to mutual blind spots between pedestrians and nearby vehicles, increasing the risk of collisions. This necessitates advanced assistive technologies to ensure pedestrian safety and improve mobility. This research proposes a Multi-modal Zebra Crossing Detection Framework for Visually Impaired Navigation (MZCDFVIN). At its core is the YOLOv11 Hough Line Transform Convolutional feature-enhanced Ship Rescue Cross-Attention (YHTCSCA) model, designed for real-time obstacle detection and zebra crossing recognition. To address training issues such as non-convergence, instability, and vanishing gradients, the Ship Rescue Optimization Algorithm (SROA) is introduced. For improved image processing and signal detection, the framework incorporates the Haze U-Net with LeCun Network Modular Fully Convolutional Network (HUNet-MC), combining Haze U-Net for image enhancement and LeNet for traffic light recognition. The system achieved high performance across evaluation metrics: 99.79% accuracy, 99.82% precision, 99.81% recall, 99.76% F1-score, 99.3% mean Average Precision (mAP), and 99.5% Intersection over Union (IoU). These results highlight the framework’s reliability and efficiency in supporting safe street-crossing for individuals with visual impairments. The combination of the YHTCSCA model with the HUNet-MC framework offers a strong solution for navigation for visually impaired pedestrians. By precisely detecting the zebra crossings and recognizing traffic signals, the system enhances street-crossing safety, reduces blind spot risks, and supports greater pedestrian independence. This multi-modal approach establishes a new benchmark for assistive navigation technologies in smart urban environments.</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

Multi-modal Zebra Crossing Detection Framework for Visually Impaired Navigation

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月9日

查看全文
  • V. M. Jemin,
  • R. Sasikumar

摘要

Independent walking poses a significant challenge for visually impaired individuals, especially during street crossings. The absence of visual cues like pedestrian lights and directional signs leads to mutual blind spots between pedestrians and nearby vehicles, increasing the risk of collisions. This necessitates advanced assistive technologies to ensure pedestrian safety and improve mobility. This research proposes a Multi-modal Zebra Crossing Detection Framework for Visually Impaired Navigation (MZCDFVIN). At its core is the YOLOv11 Hough Line Transform Convolutional feature-enhanced Ship Rescue Cross-Attention (YHTCSCA) model, designed for real-time obstacle detection and zebra crossing recognition. To address training issues such as non-convergence, instability, and vanishing gradients, the Ship Rescue Optimization Algorithm (SROA) is introduced. For improved image processing and signal detection, the framework incorporates the Haze U-Net with LeCun Network Modular Fully Convolutional Network (HUNet-MC), combining Haze U-Net for image enhancement and LeNet for traffic light recognition. The system achieved high performance across evaluation metrics: 99.79% accuracy, 99.82% precision, 99.81% recall, 99.76% F1-score, 99.3% mean Average Precision (mAP), and 99.5% Intersection over Union (IoU). These results highlight the framework’s reliability and efficiency in supporting safe street-crossing for individuals with visual impairments. The combination of the YHTCSCA model with the HUNet-MC framework offers a strong solution for navigation for visually impaired pedestrians. By precisely detecting the zebra crossings and recognizing traffic signals, the system enhances street-crossing safety, reduces blind spot risks, and supports greater pedestrian independence. This multi-modal approach establishes a new benchmark for assistive navigation technologies in smart urban environments.