(An) implementation of the real-time motion estimation in KAIST image computing system (KICS)

Abstract

A three-level hierarchical search algorithm is proposed for a real-time motion estimation in the KAIST Image Computing System (KICS). The KICS is a high-performance image-computing system with a parallel architecture, in which multiple general-purpose digital signal processors (DSP) are used. To implement the real-time MPEG-2 encoder on the KICS, a dedicated hardware for motion estimation and a fast motion estimation algorithm are developed. In addition, the software architecture of the KICS is presented for an implementation of the real-time MPEG-2 motion estimation. The performance of the KICS is evaluated in respect to the execution time of the motion estimation and the prediction part of the MPEG-2 encoding, and the simulation results of the proposed algorithm are also described. The simulation results and the performance analysis show that the proposed hardware architecture and the motion estimation algorithm can perform the MPEG-2 encoding in real-time.