| 摘 要: 针对点云配准过程易产生错误匹配对、点云配准精度低等问题,提出了一种基于三角形相似性的点云配准方法。三对正确的匹配点对在几何空间上形成的三角形一定满足相似三角形的一些性质,例如三角形三边成固定比例、三角形顶点的法线和平面法线的夹角固定。因此,可以利用这些性质有效地减少点云配准过程中错误匹配点对其影响。该方法首先计算点云中每个点的快速点特征直方图(FPFH)描述符,使用三角形相似性的方法进行特征匹配,通过奇异值(SVD)分解得到一个变换矩阵。实验结果表明,该方法与基于正态分布变换(NDT)算法与最近邻迭代(ICP)算法结合的点云配准算法相比,配准效率提升了15.3%,配准精度提升了18.2%。 |
| 关键词: FPFH;ICP;随机采样一致性;目标配准 |
|
中图分类号: TP301.6
文献标识码: A
|
|
| Point Cloud Registration Algorithm Based on Triangle Similarity |
|
ZHOU Dawei, QIAN Wei
|
(College of Mechanical Engineering, University of Shanghai f or Science and Technology, Shanghai 200093, China)
2862458500@qq.com; 1458515538@qq.com
|
| Abstract: To address issues such as wrong matching pairs and low accuracy in the process of point cloud registration, this paper proposes a point cloud registration method based on triangle similarity. Three correctly matched point pairs form triangles in geometric space that adhere to certain properties of similar triangles, such as fixed proportions of the triangle sides and a fixed angle between the normal of the triangle vertices and the plane normal. Therefore, these properties can effectively reduce the impact of incorrectly matched point pairs during the point cloud registration process. This method initially calculates the Fast Point Feature Histogram (FPFH) descriptor for each point in the point cloud, performs feature matching using triangle similarity, and obtains a transformation matrix through Singular Value Decomposition (SVD). Experimental results demonstrate that compared to the point cloud registration algorithm combining Normal Distribution Transform (NDT) and Iterative Closest Point (ICP) algorithms, the proposed method achieves a 15.3% improvement in registration efficiency and a 18.2% enhancement in registration accuracy. |
| Keywords: FPFH; ICP; Random Sample Consensus (RANSAC); target registration |
