Abstract: For the problem of camera motion parameters estimation using motion vectors, the motion vector is taken as a state, and the block location as its state index, the state evolution equation is adaptively determined according to the local correlation of regions, the mean of absolute distortion of block matching is selected as the measurement output, and thus the state space model is established.Based on the model, the motion vector is estimated by particle filter, and resampling technology is adopted to eliminate the particle degeneracy. At last, the least square method is used to determine the optimum camera parameters according to the relationship model between motion vectors and camera parameters. Experimental results show that the proposed algorithm can obtain the motion vector field of temporal and spatial consistency with less search points, and further can estimate the camera motion parameters reflecting the true camera motion.

Key words: camera parameters estimation, particle filter, motion vector, state space model, resampling

摘要： 针对利用运动矢量估计摄像机运动参数的问题，将运动矢量作为状态，以块的位置作为状态索引，根据区域的局部相关性自适应地确定状态转移方程，以块匹配的平均绝对误差作为状态估计的观测输出，从而建立状态空间模型。基于该模型利用粒子滤波器估计运动矢量，采用重采样技术来消除粒子退化的现象，最后按照运动矢量与摄像机参数的关系模型，借助最小二乘法确定最优的摄像机参数。实验表明，该算法以较少的搜索点可以估计得到具有时空一致性的运动矢量场，进而能估计出反映摄像机真实运动的摄像机运动参数。

关键词: 摄像机参数估计, 粒子滤波器, 运动矢量, 状态空间模型, 重采样