Pel Decimation Architecture for Motion Detection on Predictive Video Coding

The demand of video applications in mobile devices has grown in the last decade. However, video applications are computationally demanding tasks that can drain device battery. Since, battery capacity has not evolved at the same rate as digital electronics, is necessary to work into achieve the same tasks with less complex algorithms. The most demanding task within the video coding process is Motion Estimation and Compensation (ME/MC). Therefore, the present work aims the reduction of calls to the ME/MC procedure by introducing a power-aware scheme using approximate computing. The call reduction is achieve by introducing a simple motion detection algorithm in combination with a pel decimation technique. The reduction in calls achieves a cutback in power consumption as well as a speed-up in the coding process. The average power reduction is 91.60% while the average speed-up is 65.89% in MPEG-4. While our proposed scheme in H.264/AVC achieves an average power reduction of 93.55% and an average speed-up of 89.15%.