摘要
给出了一种新的基于神经网络的点云数据重构CAD实体模型的新方法。该方法能直接从神经网络的权值矩阵得到曲线的控制顶点、曲面的控制网格,通过神经网络的权值约束实现曲线段、曲面片之间的光滑拼接。同时对恢复的隐式表面的初始逼近网格自适应性进行优化。利用该方法恢复的网格形状,无论在表面还是在隐式曲面的轮廓部分都能获得很好的视觉效果。所有算法的时间复杂度均为O(n),可以完全实时进行。
A kind of new method for reconstructing the CAD entity model of point cloud data based on neural network was presented in this paper. This method could obtain the controlling vertex of curve / controlling mesh of curved surface directly from weight value matrix of neural network, and through the weight value restraint of neural network to realize a smooth connection between curve segments and pieces of curved surface. At the same time the self-adoption optimization was carried out on the being restored primarily approached mesh of implicit surface. Using this method, the being restored shapes of meshes would all obtain a good effect of visual sense no matter on the surface or on the profile part of implicit curved surface. The time complexity degrees of all algorithms are O（n）, and all the algorithms can be carried out completely in real-time.
出处
《机械设计》
CSCD
北大核心
2006年第4期14-15,56,共3页
Journal of Machine Design
基金
国家自然科学基金资助项目NSFC(60173046)
关键词
光顺
神经网络
反求工程
点云
fairing
neural network
reverse engineering
point cloud
