A Point Cloud Density Enhancement Method Based on Super-resolution Convolutional Neural Network

Abstract

The data acquired by airborne lidar are mainly spatial points, commonly known as point clouds. Density, as an important attribute of point clouds, is a measure of terrain fineness. The higher the density of the point cloud, the more accurate the description of the terrain and its characteristics and laws will be. In this paper, a point cloud density enhancement method based on super-resolution convolution network is proposed. Firstly, three-dimensional laser point cloud data are transformed into depth maps, then depth maps are sent to super-resolution convolution neural network for super-clarity. Finally, the super-clarity depth maps obtained by us are transformed into three-dimensional point cloud data. And we verify it through experiments. Through our method, the density of point cloud has been obviously enhanced.