基于梯度的鲁棒多尺度微运动测量算法

Robust Multiscale Gradient-based Algorithm for Micro-motion Measurement

将鲁棒估计方法与多尺度方法相结合,给出了一种基于梯度的鲁棒分层估计多尺度运动估计算法。在该算法中,多尺度金字塔的不同尺度层采用不同的梯度滤波器来估计运动。在金字塔的每一层都使用鲁棒估计,这能够较好地解决估计中由于运动不连续和图像亮度变化引起的界外值问题,从而进一步提高运动估计精度。实验模拟表明提出的算法在测量MEMS噪声图像运动时不但估计精度高,而且执行速度快。

A robust multiscale algorithm of hierarchical estimation for gradient--based motion estimation was proposed,by a combination of robust estimation method and multiscale technique. In such a multiscale approach of hierarchical estimation, the motions at each level of the pyramid were estimated by using different gradient filters. Robust estimators were applied at each level of the pyramid to overcome the problem of outliers, caused by illumination variations and motion discontinuities in motion estimation, thus the estimator accuracy was further improved. Experimental simulations show that the new algorithm provides a significant improvement in estimator accuracy,and achieves computational speedups for motion measurement of MEMS image in the presence of noise.