基于等距变换的三维点云相似性检测算法

3D point cloud similarity detection algorithm based on isometric transformation

本文研究目的是运用基于等距变换的三维点云相似性检测算法来为三维点云识别和分类问题提供新的方法.该研究方法利用投票空间的思想,认为相似的点对具有相同的等距变换.首先,通过B样条参数曲面拟合表达物体形状.其次,定义了一种主曲率和法向量组成的局部几何特征来匹配特征点对.计算点对特征之间的等距变换,将等距变换进行分类,比较同类等距下点对间特征的等距距离.最后,在每类等距变换下,对具有相同近似等距的点对进行基于PCA的聚类算法,从而得到相似点对之间构成的相似区域.实验研究结果显示在通过普林斯顿和TOSCA点云数据集下测试,对原始点云进行等距变换、噪声、降采样的处理后,能够检测到物体形状上的相似区域.研究结论:通过实验,验证了算法的可行性和鲁棒性,该方法简化了数据的预处理的过程,能够高效检测物体模型的相似性,对三维点云的分类和识别问题有着很好的应用前景.

The purpose of this paper is to propose a 3D point cloud similarity detection algorithm based on isometric transformation to provide a new method for 3D point cloud identification and classification..The research method uses theidea of voting space and thinks that similar point pairs have the same isometric transformation.Firstly,the shape of the objectis expressed by B-spline parameter surface fitting,and a local geometric feature composed of principal curvature and normalvector is defined to match the feature point pairs.Calculate the isometric trans-formation between the point-to-features,classify the isometric transformations,and compare the isometric distances of the points between the pairs of isometric points.Finally,under each type of isometric transformation,a PCA-based clustering algorithm for point pairs with the sameapproximate isometric can be used to obtain similar regions composed of similar point pairs.The experimental results showthat the results of the isometric transformation,noise and downsampling of the original point cloud are tested under thePrinceton and TOSCA point cloud da-tasets,similar areas on the shape of the object can be detected.Research conclusions:The feasibility and robustness of the method are verified by experiments. This method simplifies the data preprocessingprocess,can effectively complete the similarity of the detected object model,and has a good application prospect to the classification and recognition of 3D point clouds.