【作者】 翟欢乐；

【导师】 熊有伦； 黄永安；

【作者基本信息】 华中科技大学 ， 机械电子工程， 2009， 硕士

【摘要】 点云拼合是逆向工程的一个重要环节,其拼合精度直接影响着后面的模型重建或精度检测。目前的研究主要集中在点云的直接拼合上,利用最近点迭代算法或点云的法矢、曲率信息实现点云数据的对齐。在实际应用中,大部分零件都包含若干直线、平面、圆柱等基本几何特征,而基于几何特征进行拼合的研究较少。本文对这种包含几何特征的点云数据的拼合进行了研究,并取得了如下成果:1、给出了基于几何特征的点云拼合的约束条件,简单分析了基于这种拼合方式的测量规划。2、运用坐标变换知识对几何特征进行空间转换,并结合两种不同的评价方式确定几何特征的拼合适应度,从而建立了粗、精拼优化目标函数。3、用粒子群算法对粗、精拼优化目标函数分别给予实现,然后根据不同的收敛情况及粒子群算法的本身特点将这两种优化目标函数组装成一个优化目标函数,并用来完成发动机右半轴支架的拼合。4、对几何特征赋予参数,解决多特征情况下对应特征的匹配问题,并利用新建特征代替原有特征解决相似特征问题,从而使优化模型的适用范围具有普遍性。以机体和连杆两个实例对本文算法的优势进行研究:与点云的直接拼合相比,避免了点云的规模及重叠问题对拼合的影响,而且拼合时间短,拼合精度高;与逆向工程软件中特征对齐相比,能够自动匹配特征,避免了对应特征选择时带来的麻烦,而且拼合结果更精确。更多还原

【Abstract】 Point clouds registration, the precision of which determines the model reconstruction and the accuracy detecting, plays an important role in reverse engineering. At present, the reported work on registration mainly aims at direct registration by ICP algorithm, as well as cloud normal vector or curvature information. In actual application, most of the parts have special features, such as lines, planes and cylinders. However, there are few studies of registration based on feature. So this dissertation focuses on such registration, and gets achievements as following:1. Give the constraint condition of registration based on feature, and make simple analysis on the measure plan based on such study.2. Adopt coordinate transformation to analyze feature, and determine the fitness of registration by two different assessment modes, then establish gross and accurate registration optimizations.3. The gross and accurate registration optimizations are solved by particle swarm optimization(PSO). Then analyze their different characteristics and assembly them into one optimization to finish the registration of engine right support.4. The features are endowed parameters to solve the matching problem, when there are many constraint features in the part. This thesis also gives a way to solve the similar feature problem by constructing a new feature instead of the original one. Then the registration algorithm can be used generally.The advantages of this algorithm are investigated by two examples-engine block and connecting rod. Compared with the direct registration of point cloud, this algorithm avoids registration affects caused by the cloud scale and the overlap problem, shortens registration time, and improves registration accuracy. Compared with feature alignment in reverse software, this method realizes intelligent registration and avoids the time-consuming work of picking corresponding feature, also has higher registration accuracy.更多还原