Abstract: Point cloud registration was a key method for pose parameter measurement of large vehicle body components, but the existing algorithms were difficult to register to effective pose under a large number of abnormal point cloud interference, thereby resulting in matching distortion and inability to ensure the quality of subsequent robotic operations. To address the issue, a robust registration algorithm for vehicle body components, robust function weighted variance minimization(RFWVM) algorithm was proposed that might effectively suppress the interference of abnormal point cloud. A robust function weighted objective function was established, and the influences of abnormal point cloud in the registration processes were suppressed by applying dynamic weights that varied with the number of iterations. The rigid transformation matrix was solved iteratively by the Gauss-Newton method. The experimental results on the side walls of high-speed rail body and car door frames demonstrate that the proposed RFWVM algorithm has higher registration accuracy compared to classic algorithms, such as interactive closure point(ICP), variance minimization(VMM), weighted plus and minimum allowance variance minimization(WPMAVM), de-pseudo-weighted variance minimization(DPWVM), may effectively suppress the influences of various abnormal point clouds on registration results, and also behaves better stability and robustness. The method may effectively achieve the accurate registration of various vehicle body components.

Key words: point cloud registration, abnormal point cloud interference, robust function, vehicle body component, robotic vision measurement

摘要： 点云配准是大型车身构件位姿参数测量的关键方法，但现有算法在大量异常点云干扰下难以配准至有效位姿，从而导致匹配失真，进而无法保证后续机器人作业质量。针对此问题，提出一种能够有效抑制异常点云干扰的车身构件鲁棒性配准算法——鲁棒函数加权方差最小化(RFWVM)算法。建立鲁棒函数加权目标函数，通过施加随迭代次数可变的动态权重来抑制配准过程中异常点云的影响，并由高斯牛顿法迭代完成刚性转换矩阵的求解。以高铁白车身侧墙、汽车车门框为研究对象的试验结果表明，较经典的最近点迭代(ICP) 算法、方差最小化(VMM) 算法、加权正负余量方差最小化(WPMAVM)算法和去伪加权方差最小化(DPWVM)算法，所提出的RFWVM算法配准精度更高，能够有效抑制各种异常点云对配准结果的影响，并具有更好的稳定性和鲁棒性，能够有效实现各类车身构件点云的精确配准。

关键词: 点云配准, 异常点云干扰, 鲁棒函数, 车身构件, 机器人视觉测量