Multi-View 3D Object Reconstruction Using Coordinate Transformation
The algorithm presented in this paper takes at least two images taken by an ordinary digital camera, computes camera parameters and outputs scene geometric information. The method takes at least one right-angle triangle in the scene, install a local coordinate at each triangle, and compute camera parameters by using the local coordinate transformation. The algorithm computes orientations and displacement relationships among the triangles in the scene while correcting the previously computed camera parameter values. In order to find optimal solution, we set optimization variables as camera calibration parameters and coordinate transformation values between local coordinates. The background eliminated images and calculated camera parameters allow us to reconstruct a 3D target object in VRML. The merit of this algorithm is in handling varying focal camera, in easy initial guessing value computation and in using fewer feature points compared to Zhang’s calibration method.