K.H. Jang, Y.M. Cha, and S.K. Jung (Korea)
3D reconstruction, camera calibration, image-based modeling
The current paper deals with the 3D reconstruction problem related to uncalibrated image sequences captured from a fixed camera using a moving planar mirror. First, an efficient method is devised to estimate the intrinsic camera parameters using a set of two orthogonal vanishing points. Then, the extrinsic parameters of the virtual camera (or posture of the mirror) are computed using the two orthogonal vanishing points and a known focal length and principal point. Since the calibration results are still inaccurate due to the detection error of the two vanishing points, the extrinsic parameters are revised using corresponding corner points that are easily obtained using the initial extrinsic parameters. Finally, 3D information is computed from a set of calibrated image sequences. The proposed method is easily implemented and applicable to practical applications. Accordingly, the current paper uses a moving planar mirror to extend the reconstruction area, then an uncalibrated fixed camera captures image sequences through the moving mirror. Thus, in effect, a virtual camera moves around the object. The proposed system estimates the intrinsic camera parameters, such as the focal length and principal point, using a set of two orthogonal vanishing points. The initial extrinsic parameters of the virtual camera are then estimated by the two orthogonal vanishing points and refined by a multiple image analysis. Finally, a 3D model is built from the accurately calibrated images.
Important Links:
Go Back
