適用於動態行車視訊壓縮之H.264移動估測

摘要

動態行車視訊應用非常廣泛，例如:可作為測速影像、行車即時監控與行車紀錄器等用途。當動態行車視訊被使用在高速行駛車輛上，現今先進的H.264影像編 碼未必能達到有效的移動估測。若一段影片有受到動態行車所影響，基於動態行車視訊的特性，我們個別依畫面遠近景運算動態行車視訊因子(Speed Factor)，可更準確預測動態行車視訊移動向量，取代原H.264區塊導向的移動向量(MV prediction)，本研究所提出「修改型動態行車視訊因子移動向量(Modify Motion Vector of Speed Factor)」取代H.264在移動估測流程內「移動向量預測搜尋中心(MV Predict Search Center) 」，並依緩慢移動區塊(stationary region)的特性，結合本研究所提出「動態行車視訊因子移動向量差(Motion Vector Difference of Speed Factor)」作為區塊模式決策運算，可減少模式決策運算量並節省區塊編碼時間。本研究所提出的方法仍可相容於H.264 標準。

本研究所提出演算法(不含減少模式決策運算)，在相同實驗條件下與H.264比較，會造成PSNR提升0.018dB與bit-rate減少1.21%， 但增加16.55%編碼時間。另本研究所提出演算法(含減少模式決策運算)，在相同實驗條件下與H.264比較，此演算法會造成PSNR減少 0.001875 dB、bit-rate增加0.3%，可節省10.42%編碼時間。

　

H.264 Motion Estimation Algorithm for Moving Vehicle Video Compression

Abstract

High quality videos captured on moving vehicles are essential in many applications, such as speed-measuring, surveillance, and reconnaissance. The current state-of-art H.264 video coding may not be very effective in motion estimation when it is used in high speed moving vehicles. Based on the characteristics of moving vehicles, the motion vector can be more accurately predicted by incorporating the “Speed Factor” into the computation of the H.264 block-oriented MV prediction. This thesis proposes "Modified Motion Vector by Speed Factor" to replace" MV Prediction Search Center" in H.264. Based on characteristics of stationary region, we also propose a fast mode decision method that uses "Motion Vector Difference by Speed Factor" as block model decision criterion which can decrease model decision computation and save coding time. In addition, the proposed method is still H.264 standard compatible.

Comparing with H.264 with the same experimental setups, the proposed algorithm without fast mode reduction decision improves the image quality by increasing 0.018dB in PSNR and decreasing 1.21% in bit-rate, but increases 16.55% coding time. On the other hand, the proposed algorithm with fast mode reduction decision can saves 10.42% coding time with minor image quality degradation of 0.001875 dB decreasing in PSNR and 0.3% bit-rate increasing.