建筑物立面分类与DIM点云滤波

针对目前基于近景摄影测量方法构建建筑物立面模型过程中因密集影像匹配（DIM）点云噪声所引起的建筑物立面TIN网格模型畸变问题，本文借鉴机器学习中样本学习的思想，对建筑物立面进行了分类并对DIM点云提出了相应的滤波方法，以达到去除DIM点云噪声和改善其TIN网格模型畸变的目的。其中，针对平面结构立面，采取先对点云样本进行学习计算构建数学立面模型所需参数，再对该立面模型设定阈值并对其点云进行滤波处理的方法；针对曲面结构立面，则结合DIM点云特性先将点云样本分类标记归为立面点与非立面点，再进行样本特征值学习，使用Logistic回归算法迭代计算求解最佳回归系数，从而构建滤波分类器的方法对立面点云进行滤波处理。试验结果表明，本文滤波处理方法能将立面DIM点云噪声有效识别并去除，而且使用该方法处理后所得点云构建的建筑物立面TIN网格模型精细化程度得到有效提高，模型质量得到明显改善。

Building facade classifying and DIM point cloud filtering

Aiming at the distortion of building façade TIN grid model which caused by the dense image matching (DIM) point cloud with noise, this paper references the idea of sample learning and proposes the corresponding DIM point cloud filtering method to remove the noise point and improve the model distortion after classifying the facades. Among them, for the facade with planar structure, this paper first learns and calculates the parameters of mathematical facade model from the selected point cloud sample, and then sets the threshold for the model to filter; for the façade with curved structure, after analyzing the characteristic of the facade DIM point cloud, this paper first marks the selected point cloud sample as two types of façade point and non-facade point, then learns the feature of the marked sample and calculates the optimal regression coefficient needed for the filtering classifier by the logistic regression algorithm to remove the noise point cloud. Finally, the experimental result showed that the filtering method can effectively identify and segment the noise point cloud from the whole facade point cloud, and the facade model which built by the filtered point cloud was be refined and optimized.