Multi-resolution volumetric reconstruction using labeled regions

Abstract

The reconstruction of 3D models from camera views is becoming an element of major importance in many applications that simulate physical interaction with the real world. 3D models are amenable to access through diverse representation modalities that typically imply trade-offs between level of detail, interaction, and computational costs. We present a 3D volumetric reconstruction method that allows users to control selectively the complexity of different surface regions, while requiring only simple 2D image editing operations. An initial reconstruction at coarse resolution is followed by an iterative refining of the surface areas corresponding to the selected regions.