基于运动矢量可靠性分析的视频全局运动估计

Global motion estimation based on the reliability analysis of motion vectors

视频全局运动估计通常根据图像的运动矢量场来决定,而前景或背景中不可靠的运动矢量对运动估计的结果影响很大。为了提高估计的准确度,提出了一种基于运动矢量可靠性分析的全局运动估计算法。该算法通过对当前帧的运动矢量场进行可靠性分析,提取图像块的匹配特征并用支持向量机的方法将运动矢量分成可靠与不可靠两类,剔除不可靠的运动矢量,用可靠运动矢量建立全局运动的线性参数模型,通过线性迭代最小二乘法求解估计参数。实验结果表明:该算法相比传统方法能够更好地消除前景或背景中不可靠运动矢量的影响,特别是在采用传统方法误差很大的情况下,该算法的结果仍十分准确。

Global motion estimation of videos depends on the image motion vector field,but unreliable motion vectors affect the estimation results. A global motion estimation algorithm was developed based on the reliability analysis of motion vectors to improve the estimation accuracy. The reliability of the current frame＇s motion vector field is first analyzed,with each motion vector then classified using support vector machine method into reliable one and unreliable one according to the block＇s matching features. The reliable motion vectors are used to construct a linear parameter model with the parameters determined by iterative linear least-square estimation. Experimental results show that the algorithm effectively eliminates the effect of regions with unreliable motion vectors and improves the estimation result,even if the conventional method fails.