Projeto e gerenciamento dinâmico de uma memória de busca para a estimação de movimento na codificação de vídeo

Abstract

The high number of memory accesses during video encoding leads to a high concern related to energy consumption and battery life of mobile devices that handle digital videos. In an attempt to reduce the temporal redundancy present in neighboring frames, the ME searches for blocks of reference frames in the external memory and compares it to the current block, which is being coded. This communication between the processing unit and the external memory is numerous and causes a high energy consumption, resulting in a shorter battery life in mobile devices. Thus, considering the high number of accesses to external memory, solutions for the reduction of memory bandwidth in video coding systems should be explored. In this sense, this work performs an analysis of the distribution of the accesses of the algorithm of motion estimation Test Zone Search and, from this analysis, proposes a static memory design and a dynamic management for motion estimation from this memory. The dynamic manager splits the internal memory into two sectors and shuts off the outermost sector when it is not relevant to ME. With the sectors shutdown, this dynamic management reduces the static consumption and also decreases the number of accesses during the ME, also reducing the dynamic consumption. In this work, eleven modes of dynamic management were evaluated. With one of these modes and for a varied set of high-resolution videos, the use of memory dynamic management achieved an average of 43.59% reduction in power consumption, with negligible variation in BD-Rate of 0.49%.