(An) implementation of motion estimation/compensation for real-time MPEG-2 on KAIST image computing system (KICS)

Abstract

A real-time MPEG-2 motion estimation/compensation algorithm is implemented based on the KAIST Imaging Computing System (KICS). The KICS is developed for a multimedia platform for varies kind of data. It provides high performance and flexible features with which multimedia data is computed in real time. To implement MPEG-2 encoding, a two-level hierarchical search algorithm is proposed with a dedicated motion estimator. The implemented motion estimation algorithm is composed of two levels. First level is to compute the motion vector with double-pixel accuracy by using the dedicated motion estimator. In second level, motion vector with full-pixel accuracy is searched at nine points around the vector estimated in the first level. Four advanced DSPs of MVP used for computation tasks and a MP (RISC type processor) used for the process control of motion estimation/ compensation. In the process of implementation, various ADSP assembly language techniques are used, such as parallel, pipeline, data transfer, and so on. The simulation results show that the implemented algorithm can perform the motion estimation and compensation in real time.