Scholarlink

<p>This paper proposes a deep learning approach for 3D shoeprint point cloud classification based on an enhanced PointNet++ architecture. The model incorporates multi-scale feature extraction and attention mechanisms to improve the representation of both local and global geometric information. A structured-light scanning system was developed to capture high-precision 3D point clouds of shoeprints on soft material surfaces, followed by preprocessing to remove irrelevant background data. Using this system, a dataset of 694 samples of 3D shoeprints was collected and has been publicly released to facilitate research use. The experimental results demonstrate that the proposed model achieves 88.60% accuracy on the test set, showing excellent classification performance in 3D point cloud recognition tasks. This work demonstrates the effectiveness of combining structured-light acquisition with deep point cloud modeling for fine-grained 3D shape classification and provides a valuable reference for broader applications involving footprint analysis and unstructured 3D data understanding. The code is available at <a href="https://github.com/ZhihanTian/shoeprint_pointcloud_code">https://github.com/ZhihanTian/shoeprint_pointcloud_code</a>, and the dataset can be accessed at <a href="https://zenodo.org/records/15107266">https://zenodo.org/records/15107266</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

Enhanced 3D shoeprint classification via multi-scale PointNet++ with attention mechanisms

  • 期刊论文

  • 发表日期：

    2026年1月9日

查看全文
  • Zhihan Tian,
  • Binrui Li,
  • Linyu Huang,
  • Yong Guo

摘要

This paper proposes a deep learning approach for 3D shoeprint point cloud classification based on an enhanced PointNet++ architecture. The model incorporates multi-scale feature extraction and attention mechanisms to improve the representation of both local and global geometric information. A structured-light scanning system was developed to capture high-precision 3D point clouds of shoeprints on soft material surfaces, followed by preprocessing to remove irrelevant background data. Using this system, a dataset of 694 samples of 3D shoeprints was collected and has been publicly released to facilitate research use. The experimental results demonstrate that the proposed model achieves 88.60% accuracy on the test set, showing excellent classification performance in 3D point cloud recognition tasks. This work demonstrates the effectiveness of combining structured-light acquisition with deep point cloud modeling for fine-grained 3D shape classification and provides a valuable reference for broader applications involving footprint analysis and unstructured 3D data understanding. The code is available at https://github.com/ZhihanTian/shoeprint_pointcloud_code, and the dataset can be accessed at https://zenodo.org/records/15107266.