Constraint-based beautification and dimensioning of 3D polyhedral models reconstructed from 2D sketches

3D models reconstructed from 2D sketches are inaccurate because of the inherent inaccuracies in the input and the reconstruction method. It is therefore necessary to “beautify” them before use in CAD systems. We present a method that detects geometric constraints, such as parallel and orthogonal faces, present in the reconstructed model and then selects a subset that constrains the object sufficiently and consistently. The subset selection algorithm first prioritizes the constraints depending on their type and then uses a novel method, based on quasi-Newton optimization, to detect and eliminate redundant and inconsistent constraints. The remaining constraints then define the dimensions of the model fully and consistently. Results from our implementation show that the method can beautify and dimension recovered 3D models correctly at acceptable speed.