Representation and rendering of 3D objects or scenes have been the essential part of many applications. Image based rendering has recently received much attention as a powerful method for 3D object representation. Image based rendering considers 3D objects or scenes as a collection of images. The collection of images is taken from predefined reference viewpoints. Images at the virtual views are made by combining the provided images.
In this paper, we propose an efficient representation and rendering technique using cylinder mapping. Our system represents 3D object as DOC (depth on cylinder) and TOC (texture on cylinder). TOC and DOC are images in cylindrical space including the texture and depth information. The proposed method consists of two steps: DOC-TOC construction and virtual view synthesis. TOC is made by adding a slit of images from video sequence. DOC is made by estimating the motion of corresponding points. Given a virtual view point, virtual image is generated by an affine transform of 2D mesh. To solve an occlusion problem, TOCs and DOCs made at other position are used.
Experimental results show that the proposed method can represent 3D objects briefly using DOC and TOC and render virtual view images similar to the captured images at the real position. Also the proposed method doesn’t require any knowledge of camera information.
본 논문에서는 실린더 매핑을 사용하여 3차원 물체를 효율적으로 표현과 합성하는 방법을 제시한다. 3차원 물체는 실린더 좌표계에서 깊이와 텍스쳐 정보를 포함하는 DOC/TOC 의 두 장의 영상으로 표현된다. TOC는 다중 시점 영상의 수직방향 중심선의 집합으로 생성되고 DOC는 대응점들간의 움직임 추정을 통해 획득된다. 임의 시점에서의 영상은 TOC를 2차원 메쉬로 근사화하고 각 삼각형들을 어파인 변환하여 만들어진다. 실험 결과는 제안된 시스템이 3차원 물체를 효율적으로 표현할 수 있고 생성된 영상이 실제 획득 영상과 비슷함을 알 수 있다.
