Megumu Ueda

Real-Time free-viewpoint video generation using multiple cameras and a PC-Cluster

In this paper, we propose a system generating free-viewpoint video using multiple cameras and a PC-cluster in real-time. Our system firstly reconstructs a shape model of objects by the visual cone intersection method, secondly transforms the shape model represented in terms of a voxel form into a triangular patch form, thirdly colors vertexes of triangular patches, lastly displays the shape-color model from the virtual viewpoint directed by a user. We describe details of our system and show some experimental results.

Implementing a real-time free-viewpoint video system on a PC-cluster

In this paper, we present a system generating free-viewpoint video in real-time using multiple cameras and a PC cluster. Our system firstly reconstructs a shape model of objects by the visual cone intersection method, secondly transforms the shape model represented in terms of a voxel form into a triangular patch form, thirdly colors vertexes of triangular patches, and finally displays the shape-color model from the virtual viewpoint directed by a user. In this paper, we describe implementation details of our system and show some experimental results.

Frame rate stabilization by variable resolution shape reconstruction for on-line free-viewpoint video generation

Recently, the number of researches aiming at showing real world objects from arbitrary viewpoint have been steadily growing. The processing method is divided into three stages: 3D shape reconstruction by the visual cone intersection method, conversion of 3D shape representation from a voxel form into a triangular patch form, and coloring triangular patches. If the surface area of the object becomes larger, the frame rate decreases since the processing time of the conversion and coloring depends on the number of triangular patches. Stability of the frame rate is essential for on-line distribution of a free-viewpoint video. To solve this problem, we propose a new method which accommodates the space resolution during the 3D shape reconstruction step, thus stabilizing the number of triangular patches and the frame rate. This is achieved by raising the space resolution step by step and stopping the process on a time criteria. The reconstruction is done by using an octree-based visual cone intersection method. Experimental results show that this method makes the frame rate more stable.