【摘要】 点云数据具有大体量、高冗余且非结构化的特点。针对直接采用迭代最近点ICP算法对初始位姿较差的点云数据进行配准时存在的耗时长、鲁棒性差的问题，提出一种改进的点云配准算法。首先使用体像素网格滤波算法对点云进行精简；然后提取点云的三维尺度不变特征变换3DSIFT特征点，并结合快速点特征直方图FPFH提取特征，再利用方向向量阈值算法去除错误匹配点对，然后针对这些特征利用随机采样一致性算法RANSAC结合SVD算法计算转换参数并完成粗配准；最后采用基于KD-tree加速的改进ICP算法完成精配准。结果表明，该算法的平均配准精度为4种对比算法的17.96%、47.39%、69.88%和79.78%,并且在权衡配准精度的基础上缩短了配准时间。更多还原

【Abstract】 This paper addresses the challenges of point cloud data, which possess attributes such as large volume, high redundancy, and an unstructured nature. In light of time consumption and poor robustness issues arising when directly applying the Iterative Closest Point(ICP) algorithm to point cloud data with inadequate initial pose, an enhanced point cloud registration algorithm is proposed. First, the voxel grid filtering algorithm is used to simplify the point cloud; then extract the 3D Scale Invariant Feature Transform(3DSIFT) feature points of the point cloud, and combine with Fast Point Features Histograms(FPFH) to extract the features, next, use the direction vector threshold algorithm to remove the wrong matching point pairs, then, according to these features, the Random Sample Consensus(RANSAC) algorithm combine with the SVD algorithm is used to calculate the transformation parameters and complete the rough registration; Finally, the improved ICP algorithm based on KD-tree acceleration is used to complete the fine registration. The results show that the average registration accuracy of the proposed algorithm is 17.96%, 47.39%, 69.88% and 79.78% of the four comparison algorithms, and the registration time is shortened on the basis of weighing the registration accuracy.更多还原