Robust categorization of point cloud data

We present a low rank and sparse modeling framework and a computationally efficient algorithm for extracting Digital Terrain Models (DTMs) and foreground objects from Point Cloud Data (PCD). The model decomposes an input point cloud into three main components: bare-earth, spatially structured objects, and spatially unstructured objects or other spurious data, generating a richer output than standard bare-earth estimation algorithms. We test the proposed method using real Light Detection And Ranging (LiDAR) data.