MPEG-2压缩域中鲁棒性摄像机运动估计算法的研究

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

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。

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% ontliers. The numeric experiments show that the method proposed in this paper has satisfied efficiency and robustness in camera motion estimator.