Robust registration of virtual objects for industrial applications

Vision-based registration techniques for augmented reality systems have been the subject of intensive research recently due to their potential to accurately align virtual objects with the real world. The registration process can be done in three steps: 1) Positioning: put the virtual object at a desired position. 2) Rendering: get the 2D image of the positioned 3D virtual object 3) The rendering of the virtual object will then be merged with the image of real environment to compose an augmented image. Rendering the virtual object is not trivial work, since it not only depends on its 3D location and orientation, but also depends on the camera's location and orientation. When the camera undergoes a transformation, the parameters for rendering will also be accordingly changed. Therefore, for a realtime augmented reality system, the camera pose must be dynamically tracked at frame rate. In order to simulate the real camera, we have to know its intrinsic and extrinsic parameters. The intrinsic parameters describe camera's optical, geometric and digital characteristics. The extrinsic parameters describe position and orientation of the camera. Both intrinsic and extrinsic parameters can be estimated through a camera calibration process (for example Zonglei and Boufama's method). However, calibrating the camera in real-time, for applications similar to industry, is not necessary (for example Kutulakos & vallino's method). In this paper we present our framework based on Zonglei & Boufama's approach and we describe Kutulakos & vallino's approach (we haven't yet results).