3D reconstruction of optical markers for geometric structure modeling

• Main Authors: Gotauco, Ronald T., Layug, Lorenzo Paolo C., Rodil, Ryan Benedict F.
• Format: text
• Language: English
• Published: Animo Repository 2010
• Subjects: Computer graphics; Image processing > Digital techniques; Three-dimensional display systems; Computer Sciences; Graphics and Human Computer Interfaces
• Online Access: https://animorepository.dlsu.edu.ph/etd_bachelors/6609
• Institution: De La Salle University

This research aims to develop a system capable of reconstructing real-world objects such as cube, pyramid and diamond from 2D images captured by low-resolution web cameras to a 3D virtual model. The system is patterned after Optical motion capture systems where optical markers are placed on the corners of the object. The process of 3D reconstruction in such setups are capturing the images of the object on different views to provide disparity, calibrating the cameras to know the position and orientation of the camera as well as the internal parameters such as the focal length and principal point, correspondences from the images are then determined and a triangulation method is employed to determine the corresponding estimated 3D coordinates. However, typical systems employ manual camera calibration which consequently limits the system to a fixed camera setup. Also, such setups are expensive and require a very controlled environment such as a studio. The system tries to incorporate auto-calibration in the 3D reconstruction process to enable the movement of cameras, and determining the position and orientation of the cameras on-the-fly. The internal parameters of the camera are determined only once through manual calibration which uses calibration objects, specifically a checkered pad with boxes of known measurements, while the external parameters of the camera which are the position and orientation are determined through auto-calibration methods. Specifically, the system employs the use of the fundamental matrix which only requires the correspondences as the input data in order to be determined. Using the fundamental matrix and the internal parameters of the camera, the camera positions and orientations can easily be determined. The system is scene independent and is more dependent on the calibration of the cameras to employ 3D reconstruction. That is, given the same set camera calibration, different set of objects can be put into the scene and be reconstructed.