In most conventional systems for new viewpoint video synthesis of a moving object, calibrated multiple fixed cameras are used. Field of view of every camera must cover all the area in which the object moves. If the area is large, object size in the captured images and also in the new viewpoint video will become very small. In this paper, we propose a novel method to synthesize new viewpoint video of a moving object in natural scene, which is captured by uncalibrated moving cameras. During capture, cameras can be moved and zoomed to capture only part of a scene where there is a moving object. Projective Grid Space (PGS) which is 3D space defined by two cameras is used for object's shape reconstruction. We use SIFT for finding corresponding points in natural scene for registering moving cameras to PGS. In the experiment which is performed for demonstrating the efficacy of the proposed method, new viewpoint video is successfully synthesized from multiple moving cameras without manual operation.
展开▼