Fractal-based reconstruction of point-sets models of outdoor scenes

 

Current 3D scanning technologies, including time-of-flight scanners and multi-view techniques, enable the
acquisition of the shape and color of 3D objects, often in the form of a set of colored points. Due to surface selfocclusions
and to the limitation of the acquisition devices, acquired point clouds are often too sparse and include too
many undersampled regions to enable a high-quality rendering of the acquired scene.
This project aims at developing data structures and algorithms to provide a high-quality rendering of point-clouds
representing outdoor models. The basic idea is to reconstruct the surface in undersampled parts using fractal
construction techniques, where fractal parameters are determined locally so as to roughly approximate the original
point samples. This would enable the visualization of a much higher quality model whose overall appearance is
similar to that of the original scene.