MPEG-2压缩域中鲁棒性摄像机运动估计算法的研究

在视频检索的很多应用中,比如对象轨迹追踪,都需要首先分离摄像机运动。现提出一种MPEG-2压缩域中鲁棒性摄像机运动估计——自适应尺度残差一致性ASRC（Adaptive-Scale Residual Consensus）算法,只使用P帧的运动向量,并对多重结构噪声可达到80%的击穿点,使MPEG矢量场中奇异值的影响降到最小。对比经典LMedS估计,提出的ASRC具有更好的鲁棒性和击穿点。实验结果显示出令人满意的效果。

Research on robust estimation algorithm of camera motion in MPEG- 2 compressed domain

In many applications of video retrieval,such as object tracking,it is necessary to separate the camera motion from the object motions firstly.In this paper a robust camera motion estimation algorithm called as ASRC(adaptive-scale residual consensus) is developed.By using RANSAC and TSSE to estimate the motion vectors in P-frame in MPEG-2 video sequence,ASRC can minimize the influence of the outliers in the MPEG motion vector field.Compared with traditional method of camera motion estimation such as LMedS and M-estimator,ASRC is more robust when multiple structure-noises exist in the video and can tolerate more than 80% outliers.The numeric experiments show that the method proposed in this paper has satisfied efficiency and robustness in camera motion estimator.