融合密集匹配点云与正射影像的二维城市场景分类

2D urban scene classification based on fusion of dense matching point cloud and orthoimage

对航空摄影测量场景进行分类是摄影测童数据处理的重要环节。摄影测量匹配点云存在噪声、孔洞,影响分类精度。本文提出一种融合密集匹配点云和正射影像的城市场景分类方法。首先,分别从匹配点云、正射影像中提取出几何特征、光谱特征、纹理特征和归一化数字表面模型特征;其次,利用随机森林算法把场景中每个像素分为建筑物、植被或地面;最后,通过实验定量对比了不同的特征组合对分类精度的影响。结果表明:提取的多维特征之间存在互补关系,融合多维特征的分类能取得最优结果。

Classification of airborne photogramnietric scene is a critical step in photogrammetric data processing.Since the noise and data gaps in dense matching points cloud may affect classification accuracy,a method to classify urban scene based on the fusion of dense matching point cloud and orthoimages is proposed in this paper.Firstly, geometric,spectral,textural and normalized DSM features are extracted.Secondly,each pixel in the scene is classified as building,vegetation or ground using the random forest algorithm.Experiments are implemented to compare the results of different feature configurations.It shows that the features are complementary and classification based on the fusion of multiple dimensional features will achieve the optimal result.