Texture Synthesis-Based Hole Filling for LiDAR Data

Abstract

Light Detection and Ranging (LiDAR) scanning to produce 3D point cloud data inherently results in holes in the data due to both opaque foreground objects and transparent objects seen from only a single point of view. However, it does not suffice to simply fill the holes by interpolating the boundaries, as this produces blatant loss of structure. We present a texture synthesis-based approach to synthesizing data to fill in missing geometric data based upon the existing geometric data, and demonstrate its efficacy on large scale data sets of point cloud data of cities from the Google StreetView mapping project.