Low-Power Approximate SAD Design for Efficient Integer Motion Estimation in Video Compression
Sunitha Rani Madavaram, Mary George · Traitement du signal · 2024
Integer motion estimation (IME), which leverages the temporal redundancy in video sequences, plays a crucial role in enabling effective video compression.However, the substantial computational demands of real-time motion estimation necessitate significant processing power.This requirement poses a challenge for lightweight mobile devices, which are constrained by limited processing capabilities and rely heavily on battery power.To address this issue, an efficient design for the Sum of Absolute Differences (SAD) module has been proposed, aimed at reducing both computational complexity and power consumption in video encoding applications.The size, power, and speed of the proposed design have been evaluated and integrated using Xilinx tools.Comparative analysis indicates that the proposed design is more compact and energy-efficient than existing models.Additionally, an image sharpening approach has been employed to assess the video quality metrics, including the peak signal-to-noise ratio (PSNR) and accuracy, of the proposed and existing methods.The findings suggest significant improvements in both the performance and energy efficiency of the motion estimation process.