Sketch-Based Modeling From Single-view Drawings and Applications
To allow non-professionals or beginners to quickly create models from a set of 2D drawings, 3D reconstruction has gained attention in recent years. However, when dealing with drawings, it is necessary to treat important problems such as ambiguity and difficulty in inferring hidden parts. This is especially valid when describing characteristics of plant and animal species as explored by botanists and zoologists, where approach's have to infer or discard occluded parts. This thesis aims to reconstruct 3D objects from single-view 2D drawings, assuming that a great class of objects presents structural symmetry, or consists of multiples objects drawn as overlaid structures.
In this work, we present two frameworks, an automatic approach that leads the study and development of a robust method for the generation of 3D models using Radial Basis Functions from Hermitian data allowing to interpolate the contours of the 2D drawing with the estimated 3D normals. Despite the promising results, the use of skeletons in previous works discards details present in the drawings, creating generalized cylinders around the medial axis extracted from the drawn parts. Also, as a relevant part of the study is the development of new strategies for the segmentation of the drawings allowing to treat different parts with greater flexibility.
Furthermore, we present a interactive framework of sketch-based interface for modeling 3D objects with multiple contours as overlaid structures. Inspired by traditional illustration strategies and scientific drawings, our sketching interface allows the user to infer perceivable symmetries and occluded parts of the model prior to its automatic 3D modeling. We propose a set of 2D visual effects to enhance the visual perception of users while sketching multiple overlaid objects in single-view. Examples with drawings ranging from simple line overlapping to stripes twists illustrations and biological systematics demonstrate the capabilities of our framework.