【摘要】 车载LiDAR点云中包含地面、建筑物、行道树、路灯等丰富地物类别,自动对这些不同类别点云进行分类,对点云中目标的识别、提取及重建都具有重要意义。本文提出了一种基于Gradient Boosting的自动分类方法。该方法首先对车载激光点云进行数据预处理,然后计算点云的协方差矩阵、密度比、高程相关特征、局部平面特征、投影特征等,再计算点云特征直方图与垂直分布直方图,采用K-means方法对这两者分别进行聚类,并将其聚类类别值也作为特征,从而构建出20维的点云特征向量,应用Gradient Boosting分类方法进行自动分类。为了验证本文方法的有效性,从某城镇场景的车载激光点云数据中选取部分代表区域共144W点作为训练数据集,然后选取另一较大区域的点云共312W点作为测试数据集。使用训练好的分类器对测试数据集进行分类,分类结果总体准确率达到了93.38%,耗时631s,说明此分类方法具有较高的分类准确率,同时也具备较高的效率。更多还原

【Abstract】 Mobile LiDAR point cloud contains abundant objects such as ground, building, tree, streetlight, etc. Automatic classification of these different points is of great importance to the object recognition, extraction and reconstruction from point cloud. The paper proposed an automatic classification method based on Gradient Boosting. With the method, firstly the point cloud is pre-processed, then calculated co-variance matrix, density ratio, features derived from elevation and local plane, horizontal projection area and vertical projection area. After these calculations, the point cloud feature histogram and vertical distribution histogram are calculated and clustered respectively. Coupled with the two cluster labels, a point feature vector with a dimension of 20 could be built. Then, point cloud could be classified automatically through applying a Gradient Boosting classifier. To prove the effectiveness of the method, a train dataset(144W points) and a test dataset(312W points) are selected from a mobile laser scanning point cloud dataset. Using the trained classifier to classify the test dataset, an overall accuracy is reached as high as 93.38% with consuming 631 s. The method has a high classification accuracy as well as high efficiency when it is applied in a mobile laser scanning point cloud classification task.更多还原