Abstract: Registration remains a difficult problem in application for error estimation of complex surfaces.Different registration methods often lead to different estimation results.A proper pose is significant in precision manufacturing or in some tough tolerance situations.For this reason,a new registration method was proposed to find the registration pose in which the tolerance zone was minimized.To deal with the computational intractability,a smooth function was used to approximate uniformly the non-differentiable range function.The limited-memory Quasi-Newton algorithm was efficient and competent for large-scale and differentiable unconstrained optimization problems.The proposed method was easily implemented and worked well for dense measurement points in the premise of Iterative Closest Point (ICP) registration.Experiments on a simulation plane and a real sheet metal part were included to justify the superiority over the existing registration methods.

Key words: error estimation, registration, aggregation function, L-BFGS algorithm, minmum zone criterion

摘要： 鉴于复杂曲面误差评估问题中,不同配准方法往往导致不同的评估结果,在精度要求较高或误差形势较严峻时,恰当的配准位姿具有重要意义,提出一种面向复杂曲面误差评估的最小区域包容配准算法。为解决计算复杂性,采用凝聚函数一致光滑逼近不连续的极差函数,并利用高效有限存储拟牛顿算法求解大规模无约束非线性问题。最近点迭代算法为最小区域包容配准方法提供了收敛条件,同时加快了收敛速度,而且算法容易实施,对复杂曲面扫描数据点云有效。通过仿真平面和实测钣金件的误差评估配准实验说明了所提算法的有效性。

关键词: 误差检测, 配准, 凝聚函数, L-BFGS算法, 最小区域准则