A GPU Approach to Augmented Reality using Non-Central Catadioptric Cameras.

This paper addresses the problem of augmented reality on images acquired from non-central catadioptric systems. We propose a solution which allows us to project textured objects to images of these types of systems in real-time which, depending on the complexity of the objects, can run up to 20 fps with a reasonable resolution (near to a real-time framerate). The main contributions of our solution are related with the differences on the image formation: projection of the 3D segments onto the image of non-central catadioptric cameras; occlusions; and illumination/shading. To validate the proposed solution, we used a non-central catadioptric camera formed with a perspective camera and a spherical mirror. We used three distinct and well known objects in computer graphics: “bunny”, “happy buddha” and “dragon”, from Stanford database.