一种彩色光流场估计算法

彩色图像序列中通过求解光流梯度约束方程组进行光流估计可有效地克服孔径问题,但方程组在各等式线性相关或图像的梯度量级相对于噪声很小的像素点处得到不可靠解.为获得鲁棒估计结果,将规范化rgb颜色模型估计方法与全局平滑约束方法相结合,利用矩阵条件数判断规范化rgb方法光流解的可靠性,设定条件数阈值进行方法间切换,融合2种方法的估计结果.最后,对光流估计结果进行10次中值滤波.实验表明,该方法在获得稠密光流场的同时可有效提高光流场估计精度.

An improved algorithm for optical flow estimation from color images

An effective solution for the aperture problem in estimating flow using color image sequences is to estimate optical flow by determining the optical flow gradient constraint equations.The resulting solution may not be accurate at pixels where the equations are linearly correlated or the image gradient is lower than the noise amplitude.To improve the robustness of optical flow estimations we selectively used the normalized rgb method and the global smoothing method.With this technique,the matrix condition number of the matrix is derived in order to estimate the reliability of the outcome obtained from the normalized rgb method.Next,a threshold for condition numbers is applied to determine which method to employ.Finally,the result is filtered 10 times by a median filter.Experimental results showed that the new approach not only resolves dense optical flow fields,it also effectively improves the precision of the optical flow estimation.