【摘要】 点云配准是三维点云处理的重要基础技术，主要目标是找到源点云与目标点云之间的最佳刚性变换。然而，由于部分点云重叠率较低，配准具有一定的挑战性。在分析了裁剪迭代最近点(TrICP)算法后提出一种基于点云曲率特征相似性改进的裁剪迭代最近点算法，以提高配准精度以及鲁棒性。采用主成分分析法(PCA)构造协方差矩阵估算出点云曲率，引入kd-tree加速邻近点的搜索效率，提高效率精简点云数据的同时区别出特征点和噪声点。搜索欧几里得距离最近点时，选取皮尔逊相关系数进行曲率相似度判断，提高配准精度。该算法相较于ICP、K4PCS+ICP、SAC-IA+ICP、TrICP精度分别提高了约36.7%、50.45%、15.76%、10.7%,对比原算法迭代次数减少了34.48%,加快了迭代速度。在有噪声的情况下也优于另外几种算法。更多还原

【Abstract】 Point cloud registration is a fundamental technique in 3D point cloud processing, with the primary goal of finding the optimal rigid transformation between a source point cloud and a target point cloud. However, due to low overlap between some point clouds, registration poses certain challenges. After analyzing the trimmed iterative closest point(TrICP) algorithm, an improved trimmed iterative closest point algorithm based on point cloud curvature feature similarity is proposed to enhance registration accuracy and robustness. The principal component analysis(PCA) method is used to construct the covariance matrix to estimate point cloud curvature, and a kd-tree is introduced to accelerate the search for neighboring points, improving efficiency by simplifying the point cloud data while distinguishing feature points from noise points. When searching for the nearest Euclidean distance points, the Pearson correlation coefficient is used to assess curvature similarity, enhancing registration accuracy. Compared to ICP, K4PCS+ICP, SAC-IA+ICP, and TrICP, this algorithm improves accuracy by approximately 36.7%, 50.45%, 15.76%, and 10.7%, respectively. The number of iterations is reduced by 34.48% compared to the original algorithm, accelerating the iteration process. It also outperforms the other algorithms in noisy conditions.更多还原