A new approach with enhanced accuracy in zero motion prejudgment for motion estimation in real-time applications

摘要

For improving the accuracy and speed of motion estimation, a refined threshold estimation technique for zero motion prejudgment (ZMP) has been proposed in this manuscript. Various fixed or dynamically varying threshold techniques reported in literature treat all the blocks whose sum of absolute differences (SAD) lie below a particular threshold as stationary blocks (SB). However, in reality these blocks have high probability to be SB but not certain. Also SB with abrupt or irregular variations may not be detected with these techniques. The proposed two-level ZMP technique with its dynamic threshold at level 1 attempt to identify SB blocks with much more accuracy and precision. Level 2 of it has very high probability to identify the SB with large and abrupt distortions that are left in level 1. The proposed technique for ZMP can be applied before any fast motion estimation algorithm (FMEA). Adaptive rood pattern search algorithm (ARPS) is used as FMEA for comparing the performance of the proposed ZMP technique with other fixed and dynamic threshold techniques described in the literature. Simulation results clearly indicate that the proposed technique shows very promising results-significantly reduces the decision errors, improves the video quality in terms of PSNR and SSIM and reduces memory requirement for residual frames in comparison to fixed and dynamic threshold techniques for ZMP. Real-time applications contain sequences with higher frame rates as 30 and 60fps. The proposed technique gives promising results on sequences with higher as well as lower frame rates. Further, proposed ZMP is able to save computations by 95-99% in comparison to full search (FS) algorithm with small degradation in PSNR and SSIM. The proposed technique when embedded with fast motion estimation algorithm (ARPS) can reduce the computations up to 70% depending upon the presence of stationary blocks with negligible degradation in video quality. All this has been achieved by enhancing the accuracy in determination of stationary blocks in comparison to other ZMP techniques.