| 引用本文: | 黄天齐1 ,严 楠1,2 ,戴家树1 ,李萌阳1.基于多视角的面颈部点云配准算法研究(J/M/D/N,J:杂志，M：书，D：论文，N：报纸).期刊名称,2025，42（4）：53-61 |
| CHEN X. Adap tive slidingmode contr ol for discrete2ti me multi2inputmulti2 out put systems[ J ]. Aut omatica, 2006, 42(6): 4272-435 |
|
| 摘要: |
| 目的 为了解决非刚性配准算法在面颈部三维重建中的变形问题,设计了一种面颈部点云的采集方式,并提
出了一种相适应的局部配准再还原全局的配准算法。 方法 采集方式为深度相机分别对准头部右下颌角、左下颌角
和鼻梁 3 个位置采集点云,对采集的 3 片点云进行预处理,把点云 RGB 信息转为 HSV( Hue, Saturation, Value) 后,
定位嘴唇的位置,分割面部后留下鼻子和嘴巴区域,对其采用结合三维形状上下文特征(3DSC) 的随机采样一致性
算法( RANSAC) 进行粗配准,再使用迭代最近点算法( ICP) 进行精配准。 最后,把局部配准得到的变换矩阵应用于
原始点云上,从而得到面颈部三维点云模型。 结果 经过实验,设计的 3 个位置采集的点云能够完整覆盖整个面颈
部区域。 通过对比 5 种改进的迭代最近点配准算法,得到使用 3DSC+RANSAC+ICP 算法进行配准精度最高。 通过
设定面颈部标记点和采集不同人脸进行配准实验,对比了配准结果和真实人脸的标记点距离,结果误差均小于 2. 5
mm,验证了算法的配准精度和配准算法的鲁棒性。 结论 设计的多视角面颈部配准算法能有效配准,配准结果与真
实人脸误差小于 2. 5 mm,解决了非刚性配准算法在面颈部三维重建上的变形问题,在处理不同个体的面颈部数据
时表现出了一定的鲁棒性。 |
| 关键词: 多视角点云 刚性配准 三维重建 面颈部 |
| DOI： |
| 分类号: |
| 基金项目: |
|
| Research on Face and Neck Point Cloud Registration Algorithm Based on Multi-view Perspective |
|
HUANG Tianqi1 YAN Nan1 2 DAI Jiashu1 LI Mengyang1
|
|
1. School of Computer and Information Anhui Polytechnic University Anhui Wuhu 241000 China
2. Anhui Future Technology Research Institute Anhui Wuhu 241000 China
|
| Abstract: |
| Objective To address the deformation issues in non-rigid registration algorithms for the three-dimensional
reconstruction of facial and neck regions this paper designed a method for collecting point clouds of the facial and neck
regions and proposed an adaptive local registration algorithm that restores global registration. Methods A depth camera
was used to collect point clouds at three positions of the head the right mandibular angle the left mandibular angle and
the nasal bridge. The collected point clouds were preprocessed and the RGB information of the point cloud was converted
into HSV Hue Saturation Value . The position of the lips was located and the facial area was segmented to retain
the nose and mouth regions. The random sample consensus RANSAC algorithm combined with the three-dimensional
shape context 3DSC feature was applied for coarse registration followed by the iterative closest point ICP algorithm
for fine registration. Finally the transformation matrix obtained from local registration was applied to the original point
cloud to obtain a three-dimensional point cloud model of the facial and neck regions. Results Experimental results
showed that the point clouds collected at the three positions could fully cover the entire facial and neck region. By
comparing five improved iterative closest point registration algorithms it was found that the registration accuracy was the
highest when using the 3DSC+RANSAC+ICP algorithm. By setting facial and neck landmarks and conducting registration experiments with different facial data the registration results were compared with the distances between the marked points
on the real face with errors all less than 2. 5 mm verifying the registration accuracy and robustness of the algorithm.
Conclusion The designed multi-view facial and neck registration algorithm can effectively register with registration errors
less than 2. 5 mm compared with real faces addressing the deformation issues in non-rigid registration algorithms for three-
dimensional reconstruction of facial and neck regions and demonstrating certain robustness when processing facial and neck
data from different individuals. |
| Key words: multi-view point cloud rigid registration three-dimensional reconstruction facial and neck regions |
