Single viewpoint model completion of symmetric objects for digital inspection

摘要

The ability to create complete 3D models of real-world objects is an important task for various applications. In digital inspection, complete models allow users to analyze the entirety of an object. However, various difficulties arise for image-based acquisition techniques. First, the viewpoint planning problem must be solved. Second, each of the resulting view point captures must be combined with either zippering or 3D triangulation, both difficult problems. We observe that if an object is symmetric, then the object’s symmetry can be exploited so that a single viewpoint capture is sufficient to generate a complete, 3D triangulated model. In our work, three problems of previous approaches to generating complete models are avoided or minimized: (1) we avoid 3D triangulation, (2) we avoid searches for geometry to extend our models, and (3) we minimize viewpoint planning to the selection of a single viewpoint. Our approach also includes algorithms to mitigate global deformations due to capture error. We demonstrate our approach by capturing, reconstructing, and completing several scenes of one or more objects and illustrating several digital inspection methods with these scenes.