融合无人机LiDAR和高分辨率光学影像的点云分类方法

点云分类是激光点云数据处理的重要环节，探索自动、高效、高精度的点云分类方法具有重要意义.通过分析同机获取的LiDAR点云与高分辨率光学影像的特点，提出了融合无人机LiDAR与高分辨率光学影像的点云分类方法.首先将LiDAR点云投影到二维平面并构建不规则三角网模型，然后寻找同名点对完成与光学影像的配准与融合，进而将光学影像的光谱信息赋予无人机LiDAR点云，接着从光学影像上提取光谱特征、从LiDAR点云上提取多尺度几何特征构建分类特征集，进一步通过CFS特征选择算法实现特征集的降维，最后运用随机森林分类算法实现点云分类.实验结果表明，本文分类方法的总体精度可达89.5%，Kappa系数为0.844，与未经特征选择的分类结果相比精度提高了1.1个百分点，与单纯依靠LiDAR或者光学影像的分类相比，精度分别提高了5.4和14.9个百分点.本文方法不仅有效避免了基于点云属性内插构建新的图像融合方式带来的计算误差，同时解决了单尺度下构建几何特征时难以确定最优空间分析尺度的问题，并且对特征集进行优化选择从而有效提高了数据处理的效率.

Fusion of high-resolution optical image and unmanned aerial vehicle LiDAR for 3D point cloud classification

Point cloud classification is a critical step in the processing of LiDAR data,and exploring new automatic,efficient,high accuracy classification method is of great importance.This paper proposed a new method for point cloud classification by analyzing the feature of optical image and LiDAR data from the same aircraft.First,a TIN model was made by interpolating the LiDAR data which was projection transformed,then the registration fusion of LiDAR and optical image was achieved according to the correspondence vertexes of the two data,and the RGB attribution information from optical image was combined into LiDAR data later.Second,classification feature set was built by extracting the spectral features from optical image and multi-scale geometric features from LiDAR data.Third,a CFS feature selection method was used to reduce dimension of the classification set.Finally,a supervised classification was conducted using a random forest algorithm to classify the point cloud.Results indicate that,the overall accuracy and Kappa coefficient of the proposed method is 89.5% and 0.844,respectively.And the proposed method get an improvement in the overall accuracy by 1.1,5.4 and 14.9 percentage point when compared with no feature selection strategy,only using LiDAR data and only using optical image,respectively.The proposed method not only efficiently reduce the interpolation error when fusion based on point cloud interpolation,but also solve the problem for choosing the optimal scale to extract geometry feature in a certain analytical scale,and the data are able to be processed more efficiently when feature selection is adopted.