3D Point Cloud Upsampling for Accurate Reconstruction of Dense 2.5D Thickness Maps

This paper presents a novel robust processing methodology for computing 2.5D thickness maps from dense 3D collocated surfaces. The proposed pipeline is suitable to faithfully adjust data representation detailing as required, from preserving fine surface features to coarse interpretations. The foundations of the proposed technique exploit spatial point-based filtering, ray tracing techniques and the Robust Implicit Moving Least Squares (RIMLS) algorithm applied to dense 3D datasets, such as those acquired from laser scanners. The effectiveness of the proposed technique in overcoming traditional angular aliasing and corruption artifacts is validated with 3D ranging data acquired from internal and external surfaces of exhumed water pipes. It is shown that the resulting 2.5D maps can be more accurately and completely computed to higher resolutions, while significantly reducing the number of raytracing errors when compared with 2.5D thickness maps derived from our current approach.