面向全局坐标变换的点云配准实现及精度评价

Coordinate transformation for point cloud registration and its accuracy evaluation

三维激光扫描仪在地形测绘、数字城市、文物保护等方面具有广泛应用,但在数据处理中通常需要进行点云配准及坐标系转换。本文针对点云配准中精度要求高、计算量大等问题,提出面向全局坐标变换实现点云配准的方案,无需进行最近点搜索计算和测站之间的多次迭代计算,大大减少了计算量,且不受点云分辨率的影响;同时分别采用矩阵的LU分解和SVD分解求取仿射变换参数,并选取了10个靶标点作为检查点进行配准精度比较。实验表明在控制点少于8个时,采用矩阵的LU分解反算仿射变换参数即可获取理想的点云配准结果且算法简单易于实现;当控制点多于8个时,采用SVD分解较LU分解可以获取更高精度的点云配准精度。

Three dimensional laser scanner is widely used in the topographic surveying and mapping,cyber city,heritage protection and so on,however the point cloud registration and the coordinate transformation are always needed.The method based on global coordinate transformation is proposed for solving the problems such as high precision and large amounts of computation.The new method doesn＇t need to search for the nearest point and multiple iteration,so it reduces much computation time and it is not affected by the resolution of points.The lower and upper triangular decomposition and singular value decomposition of matrix are used for solving the affine transformation parameters.The check points are used for registration precision comparison.The experiment indicates that the good registration of point could be obtained by using the affine transformation parameters from LU decomposition of matrix when check targets are less than.The higher precision of registration could be acquired by using affine transformation parameters from SVD decomposition when check targets more than eight.