3D Algorithms
Fraunhofer Institute for Integrated Circuits
Reconstruction and Calibration
Our work in the area of 3D algorithms relates to the production of 3D film content and to the reconstruction of object shapes from images using both passive and active methods. In both contexts, exact knowledge of camera parameters is crucial to the algorithms' accuracy. The process of determining these parameters is known as camera calibration.
Camera calibration
Real-life camera systems inevitably differ in some respects from idealized camera models. Optical and other influences can create complex distortion patterns and cause other aberrations. Identifying and measuring these is part of the task of calibrating a camera system. Since camera parameters can vary during operation as a result of mechanical and thermal influences, calibration is an iterative process.
Cameras can be calibrated either intrinsically or extrinsically. The former approach involves determining camera-specific parameters such as focal length, principal point or distortion. By contrast, extrinsic calibration entails establishing either a camera's location in a spatial reference system or the relative positions of cameras in a multi-camera set-up.
Within the area of cognitive systems, we are developing easy-to-use methods for automated calibration. Another goal is to make the methods more efficient so that they can be incorporated into sophisticated compact camera systems.
3D production for film and television
Producing high-quality 3D content for the film and television market demands a perfectly calibrated multi-camera system, which usually comprises two cameras mounted on a stereo rig . In addition, a great deal of experience and familiarity with the specific requirements of 3D production is needed to optimally configure a camera. Further R&D work is carried out on methods for scene analysis as well as on supporting software for optimal configuration of camera systems.
3D reconstruction
3D models of diverse types of objects are used for a wide variety of purposes, for instance for games and special effects or in virtual worlds or medicine. Modeling real-life objects is a very time-consuming process. So-called active 3D scanners, which often use laser beams, enable quick reconstruction. However, these methods and systems are very costly, which is why our efforts in the area of cognitive systems include the exploration of passive approaches.
Provided that several views of an object or scene are available, our methods enable its three-dimensional reconstruction. This process in turn necessitates accurate calibration of the individual images and hence of the multi-camera system used. We are developing methods and algorithms for automated extraction of depth information, for generating 3D point clouds and for reconstructing surfaces from these.
While there will typically be some gaps in a reconstructed surface, complete 3D reconstruction can be achieved with the aid of a generic parameterizable 3D model.
Set Bookmark