Scholarlink

<p>Detection of obstacles holds significant importance in enhancing the mobility and the well-being of individuals with visual impairments. This paper presents YOLOv7-Plus, an enhanced variant of the YOLOv7 object detection architecture, specifically designed to address the challenges associated with small object detection in dynamic real-world scenarios. The primary objective of this enhancement is to help visually impaired individuals become more aware of their surroundings, thereby promoting greater autonomy and safety. The modifications were applied to both the backbone and head of the YOLOv7 model, resulting in a version that performs exceptionally well, particularly in environments where small and dynamic objects are prevalent. YOLOv7-Plus demonstrated superior performance in comprehensive evaluations against benchmark models such as YOLOv7, YOLOv5, FRCNN, and RCNN, achieving a mean Average Precision (mAP) of 0.87 on the WOTR dataset, which significantly outperforms the baseline YOLOv7 (0.82). Furthermore, in real-world scenarios, it consistently proved high Intersection over Union (IoU) values, ranging from 76 to 95% for critical obstacles. Beyond its technical advancements, YOLOv7-Plus demonstrates a commitment to social responsibility by leveraging cutting-edge technology to enhance accessibility and safety for visually impaired individuals. This paper not only presents a technical enhancement to object detection but also emphasizes the profound societal impact, particularly in providing enhanced accessibility for unsighted individuals.</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

Real-Time Object Detection for Visually Impaired People Using an Improved YOLOv7-Plus Architecture

  • 期刊论文

  • 发表日期：

    2025年11月5日

查看全文
  • Loubna Bougheloum,
  • Mounir Bousbia Salah,
  • Maamar Bettayeb

摘要

Detection of obstacles holds significant importance in enhancing the mobility and the well-being of individuals with visual impairments. This paper presents YOLOv7-Plus, an enhanced variant of the YOLOv7 object detection architecture, specifically designed to address the challenges associated with small object detection in dynamic real-world scenarios. The primary objective of this enhancement is to help visually impaired individuals become more aware of their surroundings, thereby promoting greater autonomy and safety. The modifications were applied to both the backbone and head of the YOLOv7 model, resulting in a version that performs exceptionally well, particularly in environments where small and dynamic objects are prevalent. YOLOv7-Plus demonstrated superior performance in comprehensive evaluations against benchmark models such as YOLOv7, YOLOv5, FRCNN, and RCNN, achieving a mean Average Precision (mAP) of 0.87 on the WOTR dataset, which significantly outperforms the baseline YOLOv7 (0.82). Furthermore, in real-world scenarios, it consistently proved high Intersection over Union (IoU) values, ranging from 76 to 95% for critical obstacles. Beyond its technical advancements, YOLOv7-Plus demonstrates a commitment to social responsibility by leveraging cutting-edge technology to enhance accessibility and safety for visually impaired individuals. This paper not only presents a technical enhancement to object detection but also emphasizes the profound societal impact, particularly in providing enhanced accessibility for unsighted individuals.