基于三参数模型的快速全局运动估计

Fast Global Motion Estimation Based on 3-Parameter Global Motion Model

提出了一种新的全局运动估计方法———基于三参数模型的快速全局运动估计.新的参数模型在保证准确性的同时,使用更少的参数来描述和估计全局运动,从而简化了计算复杂度.此外,对光流场计算做出了两方面改进:(1)提出了宏块预判的方法,计算光流场前对宏块的梯度信息进行预分析,通过减少参与计算的宏块数目提高光流场的计算速度;(2)提出了快速估计宏块运动向量的方法,在块匹配的过程中同时考虑图像的梯度和灰度信息,通过引入更多的约束提高运动向量的计算速度.实验证明了该方法的有效性.

Global motion estimation is an important task for many video processing applications. In this paper, a new fast method is proposed to estimate global motions, which is based on a 3-parameter model. While the new model uses less parameter to describe and estimate global motion, its estimation results are still accurate adequately. Besides, another two approaches are proposed to accelerate the procedure of motion field calculation： First, an approach for block judgment is presented. Before motion field calculation, gradient information is analyzed to judge the blocks, by which the number of blocks for subsequent calculation is reduced. Second, a fast algorithm is proposed to compute optical flow, based on both gradient information and density information. Extensive experiments show the effectiveness of this technique.