Area-efficient fast scheduling schemes for MVC prediction architecture

Abstract

While multi-view video system offers users various three-dimensional scenes, its hardware architecture requires more power consumption, chip size and processing time. Each view in Group of Group of Pictures (GoGoP) has different frame structure and different number of reference frames. The multi-view video coding (MVC) performance heavily relies on the frame scheduling architecture for motion estimation (ME) and disparity estimation (DE). Therefore, we need to develop efficient frame scheduling schemes for MVC. In this paper, we propose two frame scheduling schemes: hardware resource aware scheduling (HRaS) and buffering time aware scheduling (BTaS). In the proposed scheduling schemes, the GoGoP is represented by a graph in which edges represent relationship between reference frames and current frames. HRaS relocates frames in GoGoP so that total hardware resources for MVC prediction are reduced. BTaS relocates the frames based on the number of edges so that prediction time and buffering time of the frames are reduced. Through experimental results, we verified the efficiency of the proposed scheduling architectures in terms of power dissipation, area, MVC prediction time, and buffering time.