Scholarlink

<p>Corner point detection is a crucial step in camera calibration and fish-eye distortion correction. However, traditional corner detection algorithms rely on localized feature extraction, which is prone to false detections, thereby compromising the accuracy of camera sensor pose estimation. To address this issue, this paper proposes a corner point detection method based on deep learning, designing a novel loss function and employing the BiFPN structure to enhance the performance of the YOLOv8 model in corner point detection. Experimental results demonstrate that both the loss function and BiFPN structure significantly improve the YOLO model, with the proposed YOLOv8-BiFPN model showing 3% increase in precision, 5% increase in recall, and 2% increase in mean Average Precision (mAP50). Finally, pixel-level comparison experiments between the proposed YOLOv8-BiFPN model and three traditional corner detection methods are conducted. The results indicate that the proposed YOLOv8-BiFPN model significantly outperforms the traditional methods in corner point detection, with no false detections and an accuracy of 91.01%. We have made public the full source code of the proposed algorithms in the paper. The source code is at <a href="https://github.com/baozixixixi/corner-detection">https://github.com/baozixixixi/corner-detection</a>.</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

Checkerboard corner point detection for enhanced accuracy in fish-eye camera images

  • 期刊论文

  • 发表日期：

    2025年7月24日

查看全文
  • Kun Zhou,
  • Jiacai Liao,
  • Xiaoli Liu,
  • Lin Hu,
  • Jinlai Zhang,
  • Kai Gao

摘要

Corner point detection is a crucial step in camera calibration and fish-eye distortion correction. However, traditional corner detection algorithms rely on localized feature extraction, which is prone to false detections, thereby compromising the accuracy of camera sensor pose estimation. To address this issue, this paper proposes a corner point detection method based on deep learning, designing a novel loss function and employing the BiFPN structure to enhance the performance of the YOLOv8 model in corner point detection. Experimental results demonstrate that both the loss function and BiFPN structure significantly improve the YOLO model, with the proposed YOLOv8-BiFPN model showing 3% increase in precision, 5% increase in recall, and 2% increase in mean Average Precision (mAP50). Finally, pixel-level comparison experiments between the proposed YOLOv8-BiFPN model and three traditional corner detection methods are conducted. The results indicate that the proposed YOLOv8-BiFPN model significantly outperforms the traditional methods in corner point detection, with no false detections and an accuracy of 91.01%. We have made public the full source code of the proposed algorithms in the paper. The source code is at https://github.com/baozixixixi/corner-detection.