Calibration of multiple cameras for a turntable-based 3D scanner
In a method and system for calibrating a multiple cameras, a calibration object is placed on a turntable while a set of images is acquired with each camera while rotating the turntable. Vertices of a calibration pattern are located in each set of images, and intrinsic parameters for each camera are determined from the vertices. An orientation of a rotation axis of the turntable relative to a position of each camera is estimated from the intrinsic parameters of each camera. A dependency of each orientation on a global geometry of the calibration pattern is minimized. Then, one camera is selected as a fixed camera, all other cameras are designated as floating cameras. A translation vector between the fixed camera and each floating camera is determined, and the translation vectors are used to place the cameras in a common coordinate frame according to the translating vectors.
1. A method for calibrating a plurality of cameras with a calibration object placed on a turntable, comprising:
acquiring a set of images with each camera of the calibration object while rotating the turntable;
locating vertices of a calibration pattern in each set of images;
determining intrinsic parameters for each camera from the vertices;
estimating an orientation of a rotation axis of the turntable relative to a position of each camera from the intrinsic parameters of each camera;
minimizing a dependency of each orientation on a global geometry of the calibration pattern;
selecting one camera as a fixed camera and designating all other cameras as floating cameras;
determining a translation vector between the fixed camera and each floating camera; and
placing the cameras in a common coordinate frame according to the translating vectors.
2. The method of claim 1 wherein the calibration object is a cube, and the calibration pattern is a rectangular bar on each face of the cube.
3. The method of claim 1 wherein the locating further comprises:
detecting edges in the calibration pattern; and
iteratively fitting line segments to the edges to locate the vertices.
4. The method of claim 3 wherein the line segments are fitted by orthogonal regression.
5. The method of claim 2 wherein each rectangular bar has a different color to automatically track the calibration pattern in the sets of images.
6. The method of claim 1 wherein a plane is fitted to focal points of the intrinsic parameters to estimate the orientation of the rotation axis.
7. A system for calibrating a plurality of cameras with a calibration object, comprising:
a turntable for placing a calibration object;
a plurality of cameras for acquiring a set of images with each camera of the calibration object while rotating the turntable;
means for locating vertices of a calibration pattern in each set of images;
means for determining intrinsic parameters for each camera from the vertices;
means for estimating an orientation of a rotation axis of the turntable relative to a position of each camera from the intrinsic parameters of each camera;
means for minimizing a dependency of each orientation on a global geometry of the calibration pattern;
means for selecting one camera as a fixed camera and designating all other cameras as floating cameras;
means for determining a translation vector between the fixed camera and each floating camera; and
means for placing the cameras in a common coordinate frame according to the translating vectors.