3DMRRC: 3D Mesh Reconstruction using Ray Casting for stationary subjects

Abstract

With increasing processing power and better rendering capabilities, it is now feasible to view a 3D mesh of millions of triangles. It has various application domains such as augmented preview, medical field, construction and many more. However, creating such a detailed mesh requires manually modelling each feature in a tool (like Blender). A quicker approach would be to use photogrammetry that requires a set of cameras capturing each visual aspect of the subject from specific angles. As we can observe, both of these techniques are resource intensive (time and physical camera setup respectively). Further, Deep learning based methods have been proposed, but these do not work on novel geometry. We propose a methodology that makes use of the unique approach of ray casting which takes image sequences as an input and produce a 3 dimensional mesh. This approach can work for any novel geometry since no pre-learning is being performed as well as the time taken would be far less than manual modelling and also works with a single camera setup.