Abstract: This paper presents a camera tracking algorithm based on the statistics of feature blocks，it can be used to track the movement of camera in video files.First，spilling N random points in the region of video，finding out the block that having maximal differences of color nearby the random point.Then the authors analyzes adjacent frame to find the new position of the feature block that best matched.According to the movement of all blocks，the authors seek the mean and eliminate the blocks which having excessive variance of movement，using the remaining blocks to do the statistics.Last，the authors figures out the parameters of linear transformation equation by the Least-squares method.Experiments show that the algorithm achieves more accurate and robustness tracking effect with various types of video files in ordinary PC.

Key words: camera tracking, camera calibration, feature blocks, random statistic, projection transform

摘要： 提出一种基于特征块统计的摄像机跟踪算法，可用于视频中摄像机运动的快速跟踪和定位。首先在视频范围内抛洒N个随机点，在随机点周围一定区域内寻找出最具颜色差异的像素块作为特征块，然后分析相邻帧中与此特征块最佳匹配的新位置，根据各块的移动情况求出均值，再去掉移动方差过大的特征块，保留余下的特征块进行统计。用最小二乘法统计方法求出连续视频帧之间运动参数的线性变换方程。实验表明，该算法具有较好的跟踪检测效果，对各类视频文件具有较好的鲁棒性，在普通PC机中实现了较为精确的摄像机运动跟踪检测。

关键词: 摄像机跟踪, 摄像机标定, 特征块, 随机统计, 投影变换