基于模型的自由形态物体匹配方法研究  

Model\|based 3D free\|form object registration

作　　者：方昭江[1,2] 丁明跃[1,2] 

机构地区：[1]华中理工大学图像识别与人工智能研究所 [2]图像信息处理与智能控制国家教育部开放实验室

出　　处：《红外与激光工程》1999年第3期5-8,共4页Infrared and Laser Engineering

基　　金：国家教委跨世纪优秀人才基金;国家自然科学基金

摘　　要：匹配是物体识别中的一个重要部分。针对自由形态物体识别问题提出了一种快速匹配算法以计算模型与物体之间的刚体变换。算法在最近点迭代匹配算法的基础上，针对用结构光方法重建获得的三维物体表面，对数据点进行了加权处理，以减弱曲面重建产生误差对匹配精度的影响。并对匹配运算采用了多分辨率分层匹配方法，对低分辨率模型的刚体变换进行粗略计算，将结果作为高分辨率模型数据计算的初始值，大大减少了匹配的计算代价。实验结果证明了文中提出方法的有效性。Registration is an important problem in object recognition. In this paper, a new algorithm to accelerate the speed of 3D free\|form object registration is proposed. The algorithm determines rigid transformation between model and data. Our method is based on the iterative closest point (ICP) algorithm, but takes a membership of weighted value to points in 3D surface obtained by reconstruction with structure light approach to weaken the influence of reconstruction error. With the approach of hierarchical model construction, rough rigid transformation for low resolution model is computed and introduced as the initial configuration of registration for higher resolution. Cost of registration computation is highly reduced. The experiments demonstrate the feasibility of approach.\=\;

关 键 词：计算机视觉 自由形态物体 匹配 模式识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]