Scholarlink

The identification of outdoor scene text proves difficult because outdoor environments feature complex backgrounds with dynamic lighting conditions and partial text obscuration. The proposed framework operates in two stages to identify numbers through a DDBB database that was specifically designed for outdoor sporting and racing settings. The system combines YOLOv8 nms variants with s and m versions to detect bib numbers and utilizes MMOCR toolkit for precise text recognition. Research trials were done through 60 and 120 training epoch procedures with precision, recall, and mean Average Precision (mAP) serving as performance parameters. The YOLOv8m model reached the highest mAP@0.5 score of 0.976 alongside an mAP@0.5:0.95 score of 0.712 while running for 60 epochs and attained 0.971 and 0.707 at 120 epochs illustrating its excellence in accuracy and stability. The proposed DDBB database and YOLOv8-MMOCR pipeline demonstrates exceptional capability in addressing real-world outdoor text detection needs through numbers detection in dynamic environments.

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

Dual-Branch Scene Text Detection for Robust Competition Number Identification in Dynamic Outdoor Environments

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Virendra Singh Solanki,
  • Awanit Kumar,
  • Snehlata Ajmera,
  • Nirmal Singh,
  • Sheshang Degadwala,
  • Dhairya Vyas

摘要

The identification of outdoor scene text proves difficult because outdoor environments feature complex backgrounds with dynamic lighting conditions and partial text obscuration. The proposed framework operates in two stages to identify numbers through a DDBB database that was specifically designed for outdoor sporting and racing settings. The system combines YOLOv8 nms variants with s and m versions to detect bib numbers and utilizes MMOCR toolkit for precise text recognition. Research trials were done through 60 and 120 training epoch procedures with precision, recall, and mean Average Precision (mAP) serving as performance parameters. The YOLOv8m model reached the highest mAP@0.5 score of 0.976 alongside an mAP@0.5:0.95 score of 0.712 while running for 60 epochs and attained 0.971 and 0.707 at 120 epochs illustrating its excellence in accuracy and stability. The proposed DDBB database and YOLOv8-MMOCR pipeline demonstrates exceptional capability in addressing real-world outdoor text detection needs through numbers detection in dynamic environments.