Motion estimation algorithm for real-time systems

Abstract

The paper presents an algorithmic enhancement of the full-search block-matching algorithm for motion estimation for real-time systems. The multi-stage interval based motion estimation (MIME) algorithm reduces the computational load by successively eliminating candidate blocks from the search window. The elimination process uses low bit-resolution and it is applied in multiple stages for motion vector computation. On average, MIME eliminates more than 88% of the candidate blocks in the search window after the first and second stage. Based on these results, in a real-time environment, the algorithm can be stopped at any stage without incurring significant loss in motion estimation accuracy. Simulation results show that, in the worst case scenario when the algorithm stops after first stage, there is an average loss of only 3 dB in PSNR as compared to the full-search block-matching algorithm and an average loss of 1.2 dB if the algorithm is stopped after the second step.