文摘
The architectures of image acquisition and computation usually play important roles in most computer vision systems, especially in the multi-camera dynamic light-field acquisition and rendering systems for virtual reality. This paper designs a general distributed stream computing architecture to support light-field data stream acquisition and computation. Taking advantage of the distributed computing framework and in-memory high-speed processing engine, this architecture combines stream and batch processing together, which could reduce the computation burden. In order to evaluate the performance of proposed distributed stream computing architecture, we construct a dynamic light-field acquisition and rendering system, and the light-field data are obtained from a 10-meter-diameter and 7-meter-height hemispherical steel-frame dome that is equipped with 20 cameras and 2000 LED lightings. Experiments results show our system can continuously acquire and render light-field data at more than 1 frame per second with limited computational resources.