GPU_CPU based parallel architecture for reduction in power consumption

Real-time and power management are required for high quality video in mobile environment. This paper presents a GPU(graphics processing unit) based parallel architecture for Multi-view Video (MVC) decoder which reaches these requirements. The 3D video based on stereo or multi-view representation is becoming widely popular. Real-time decoding of such video is an important concern as the number and spatial/temporal resolution of views increase. Significant improvement in video compression capability has been demonstrated by using H.264/Advanced Video Coding (AVC) standard. Multi-view video (MVC) is the extension of H.264/ AVC. In this paper, we proposed MVC decoder architecture based on parallel combination of Cortex-A8 processor and GPU (graphics processing unit). The basic operations are performed by the processor while the motion compensation (MC) feedback loop of the decoder is moved to GPU in order to achieve decoding efficiently. The experimental results show that compared to general implementation, the proposed parallel processing of a particular task in an embedded system can reconstruct the target images with higher quality with reduced processing time and energy saving with almost the same compression performance.