Three-Dimensional Object Modeling From Uncalibrated Image Sequences

摘要

A method to recover 3d metric model of a rigid object from multiple perspective views with uncalibrated camera is presented. No prior knowledge about the object shape and the camera parameters are required. The method composes of three steps. First, the internal camera parameters are calibrated through an reference object. Second, an initial 3D model is acquired from two views. In the third step, the model computed in the last step is used to estimate the motion of the other views. A robust point tracking algorithm and a Kalman Filter framework is used in this step. The results on synthetic imagery illustrate that the proposed scheme is robust and accurate.