School of Information, Xi’an University of Finance and Economics, Xi’an 710100, China
Show less
文章历史+
Received
Published
Revised
2025-01-06
2026-02-18
2025-10-27
Online Date
2026-05-27
PDF (27463K)
摘要
针对当前点云配准算法容易受到初始位姿、噪声的影响,导致点云配准精度较低及耗时较长等问题,提出一种基于纹理和几何特征的点云配准算法。首先,提取点云的曲率、密度等几何特征以及含有颜色信息的纹理特征,筛选出特征点集;然后,利用FPFH-3DLGOP特征描述符对特征点集进行描述,并通过随机采样一致性算法(random sample consensus,RANSAC)进行关键点匹配,完成点云粗配准,从而得到一个良好的初始位姿;最后,使用基于KD树的迭代最近点算法(iterative closest point, ICP)实现点云精配准。实验采用斯坦福数据集及Sketchfab公开数据集进行算法验证,实验结果表明,所提算法对比已有算法在初始位姿较差及含噪声情况下时间效率和精度上分别提升了约25%和30%。可见在相同条件下,所提算法相比已有算法具有更高的时间效率和精度。
Abstract
Point cloud registration algorithms are susceptible to initial pose and noise, resulting in low accuracy and increased processing time. To address these issues, a point cloud registration algorithm based on texture and geometric features was proposed. First, geometric features such as curvature and density, along with texture features containing color information, were extracted from the point cloud to select a set of feature points. Then, the feature point set was described using FPFH-3DLGOP descriptors, and key points were matched through the random sample consensus (RANSAC) algorithm to achieve coarse registration, thereby obtaining a favorable initial pose. Finally, fine registration was performed using the iterative closest point (ICP) algorithm based on a KD-tree. Experiments were conducted using the Stanford dataset and publicly available Sketchfab data to validate the algorithm. The experimental results show that the proposed algorithm improves time efficiency and accuracy by approximately 25% and 30% respectively, compared to existing algorithms under conditions of poor initial pose and noise. It is concluded that, under the same conditions, the proposed algorithm offers higher time efficiency and accuracy compared to conventional algorithms.
ZHAO Fu-qun, LI Zhi-hao, YU Jia-le.
Three-dimensional Point Cloud Registration Algorithm Based on Color Texture and Geometric Features[J].
Science Technology and Engineering, 2026, 26(5): 2051-2059 DOI:10.12404/j.issn.1671-1815.2500131
当前应用最广泛且最经典的配准算法是Besl等[4]提出的迭代最近点(iterative closest point, ICP)算法,该算法相对简单,但对初始配准位置要求较高,如果初始位置不佳,容易陷入局部最优,无法满足精确配准的需求。为了解决这个问题,国内外学者针点云配准流程和场景提出了各种改进算法。戴静兰等[5]通过寻找点云的主方向得到良好的初始位置,并根据点云的曲率特征寻找对应特征点,提高了算法运行的效率,但对于曲率特征不明显的物体,配准结果较差。林森等[6]提出了一种基于三尺度特征的点云配准方法,通过融合曲率变化、测量角度和特征值性质指标,显著提高了配准的精度,但该方法获取多尺度特征点描述符参数较多,增加了数据计算的复杂性。白创等[7]提出了一种基于PVDAC描述子的RGB-D三维点云配准方法,提高了算法运行的稳定性,但在针对单一物体多视角多帧点云配准时,物体周围存在的噪声会使得点云粗配准的效果变差。王勇等[8]针对三维彩色物体的配准问题,利用几何和颜色信息构造混合特征,在精度和纹理保持效果上相比ICP等配准算法有所提升,但所用颜色信息为单通道的灰度值,对其描述性能有一定影响。Li等[9]提出了一种利用点的RGB值进行三维点云配准的快速方法,保证准确性的同时提高了效率,但对于物体颜色变化不明显区域的配准效果较差。Zhu等[10]提出了一种新的基于曲面局部变化的关键点检测方法,提高了配准的精度,但在配准过程中的耗时仍存在不足。Kang等[11]提出了一种基于局部几何特征约束结合内层和重叠比综合评价的配准方法,保证了准确性的同时提升了算法的鲁棒性,但在配准过程中会消耗大量的时间。Tombari等[12]设计了CSHOT描述符,通过将颜色信息融入SHOT形成新的特征描述符,颜色信息的加入虽然提高算法配准精度,但却会导致描述符的计算更加复杂、维度增高以及配准的时长变长。Li等[13]通过融合几何和彩色信息进行特征提取,但色彩信息的特征描述算法大多复杂度高,当点云数量较多时,特征提取和匹配的效率不高,用于彩色点云配准任务时耗时过长。
ShenChen, XueLong, LiWanqi, et al. 3D point cloud registration method for robot disassembly targets based on KD-ICP algorithm[J]. Science Technology and Engineering, 2024, 24(21): 9025-9032.
KongLinghui, LuDezhong, YeFei. Application of 3D laser scanning technology in historical building facade surveying[J]. Bulletin of Surveying and Mapping, 2022(8): 165-168.
HuXiaojing, LiuYu, ChenYanbo, et al. Exploration of urban 3D modeling methods integrating multi-source data[J]. Bulletin of Surveying and Mapping, 2023(11): 128-131.
[7]
BeslP J, McKayN D. Method for registration of 3-D shapes[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1992, 14(2): 239-256.
LinSen, ZhangQiang. Point cloud registration using neighborhood point Information for description and matching[J]. Optics and Precision Engineering, 2022, 30(8): 984-997.
BaiChuang, ChenLi, YanYu. An RGB-D 3D point cloud registration method based on PVDAC descriptor[J]. Journal of Hunan University (Natural Sciences), 2023, 50(2): 95-101.
WangYong, LiChun. Adaptive point cloud registration algorithm based on color and geometric information[J]. Advances in Laser and Optoelectronics, 2020, 57(20): 146-154.
[16]
LiP, WangR, WangY, et al. Fast method of registration for 3D RGB point cloud with improved four initial point pairs algorithm[J]. Sensors, 2019, 20(1): 138.
[17]
ZhuJ, HuangZ, YueX, et al. Point cloud registration based on local variation of surface keypoints[J]. Electronics, 2023, 13(1): 35.
[18]
KangC, GengC, LinZ, et al. Point cloud registration method based on geometric constraint and transformation evaluation[J]. Sensors, 2024, 24(6): 1-28.
[19]
TombariF, SaltiS, Di StefanoL. A combined texture-shape descriptor for enhanced 3D feature matching[C]// 2011 18th IEEE International Conference on Image Processing. Brussels: IEEE, 2011: 809-812.
[20]
LiY, YinzhengL, XiaG, et al. Point cloud classification algorithm based on the fusion of the local binary pattern features and structural features of Voxels[J]. Remote Sensing, 2021, 13(16): 3156.
[21]
LiuQ, ZhuH, ZhouY, et al. Density-invariant features for distant point cloud registration[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Paris: IEEE, 2023: 18215-18225.
[22]
WangG, GaoX Y, ZhangT Z, et al. LiDAR odometry and mapping based on neighborhood information constraints for rugged terrain[J]. Remote Sensing, 2022, 14(20): 5229.
[23]
YangB, HuangZ, LiY, et al. Hybrid3D: learning 3D hybrid features with point clouds and multi-view images for point cloud registration[J]. Science China Information Sciences, 2023, 66(7): 172101.
LuJun, LiuShuang, QiaoPengfei, et al. Point cloud registration integrating color and geometric information[J]. Journal of Beijing Institute of Technology, 2024, 44(4): 428-438.
LiYanhong, YanJianguo, WangXiaoyan. Improved iterative closest point algorithm for LiDAR target point cloud registration based on neighborhood curvature[J]. Advances in Laser and Optoelectronics, 2023, 60(2): 396-401.