LiDAR Super-Resolution Based on Segmentation and Geometric Analysis

In this work, we propose a new light detection and ranging (LiDAR) super-resolution method for indoor and outdoor scenes in urban environments without training, converting to images and auxiliary sensors. To generate high-resolution (high-res) point clouds from the low-resolution (low-res) measurement of a sparse LiDAR, we deconstruct the whole problem into three sequential modules. First, to avoid feature interference, the raw data are segmented into the ground and nonground points. Then, these two types of points are reorganized into a more regular representation of the coordinates. Finally, the high-res point clouds are generated via a new geometric method. We present abundant synthetic and real data based on testing and evaluation of the proposed method. Qualitative and quantitative comparisons show that our method is effective and robust.