Current Issue Cover
一种基于皇后模板抽样的全局运动估计方法

樊玉平1, 田裕鹏1(南京航空航天大学自动化学院,南京 210016)

摘 要
在目标跟踪、视频编码等领域都需要对全局运动进行快速准确的估计。针对既有摄像机运动又有前景物体运动的视频图像序列的全局运动估计问题,提出了一种新的全局运动估计算法。该新算法先采用3层金字塔进行多分辨率计算;然后在每层运算中都采用基于块的去外点法来有效去除外点。为了提高运算速度,在金字塔下面两层中还分别采用四皇后、八皇后模板抽取背景像素,只有被抽到的像素才参与迭代计算。通过对比试验可以看到,该算法在不降低精度的情况下,大大提高了估计速度,达到了实时性要求。
关键词
Global Motion Estimation Algorithm Based on Queen Patterns Subsample

()

Abstract
Fast and accurate estimation of global motion is necessary in target tracking,video coding and etc.Considering the global motion estimation of video with both camera motion and foreground objects motion,a new global motion estimation algorithm is proposed.In the algorithm a three-level pyramid is utilized.Then outliers are effectively eliminated with the block-based methods in the calculation of each pyramid level.In order to accelerate the algorithm,4-Queen and 8-Queen patterns are also used to sample background pixels respectively on the intermediate level and the base level,and only the extracted pixels are contributed to the iterative calculation.The experimental results show that the algorithm improved the estimation speed greatly without reducing the accuracy and satisfied the real-time request.
Keywords

订阅号|日报