基于ISS和改进的FPFH特征的NDT点云配准算法研究  

Research on NDT Point Cloud Registration Algorithm Based on Iss and Improved FPFH Features

作　　者：甘鑫斌 赵德鹏 刘永生 

机构地区：[1]长安大学工程机械学院,陕西省西安市710064

出　　处：《工程机械》2025年第5期10-14,I0001,共6页Construction Machinery and Equipment

摘　　要：在点云配准过程中,针对正态分布变换(NDT)算法在处理大规模点云数据时存在配准效率较低和精度依赖初始位姿的问题,提出一种基于内部形状描述子(ISS)和改进的快速点特征直方图(FPFH)的NDT点云配准算法。该算法通过结合特征点与非特征点配准方法,分为粗配准和精配准两个阶段,以实现不同条件下点云数据的高精度对齐。在粗配准阶段,首先利用ISS算法提取关键点并计算FPFH特征,然后通过特征匹配与奇异值分解(SVD)方法计算初始变换矩阵。在精配准阶段,采用NDT算法,通过局部区域的正态分布建模,优化源点云与目标点云之间的相对变换。试验结果表明,该算法在配准精度和计算效率上均优于传统ICP算法,特别是在复杂场景和噪声环境下展现了更强的鲁棒性。In the process of point cloud registration,in viewof theproblems of low registration efficiency and accuracy dependenceon the initial pose when the normal distributions distributions transform(NDT)algorithm processes large-scale point cloud data,an NDT point cloud registration algorithm based on intrinsic shape signatures(ISS)and improved fast point feature histograms(FPFH)is proposed.The algorithm combines the feature point with non-feature point registration methods and is divided into two stages,coarse registration and fine registration,to realize high-precision alignment of point cloud data under different conditions.In the coarse registration stage,the ISS algorithm is first used to extract the key points and calculate FPFH features,and then the initial transformation matrix is calculated by feature matching and singular value decomposition(SVD)methods.In the fine registration stage,the NDT algorithm is used to optimize the relative transformation between the source point cloud and the target point cloud by modelling the normal distribution of local regions.The experimental results show that,the algorithm outperforms the traditional ICP algorithm in terms of registration precision and calculation efficiency and presents stronger robustness especially in complex scenes and noisy environments.

关 键 词：ISS特征点 FPFH特征 正态分布变换 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]