利用相关函数估计在图象光流计算中的误差

从光流来决定照相机与它所见的场景之间的相对运动，是一个困难的问题。不少人希望计算的误差能在一种可接受的范围内。然而，大部分光流研究用的是基于亮度不变假设的模型，而这种假设在不少情况下并不适用。当使用允许图象亮度改变的模型时，误差或者说偏差在图象光流计算中的性质，并利用一个各向同性的图象亮度的空间相关函数，估计在图象光流计算里的误差（偏差）。研究结果是：１）当亮度改变时，除非运动为零，在图象光流中存在误差；２）对较随机的图象来讲，图象光流中的误差比光滑图象中的误差更大；３）利用定义的相关函数，可能定量地估计在图象光流计算中的误差。

Using correlation function to estimate errors in optical flow computation

elative motion of a camera with respect to its imaged scene can be determined from the variation of image brightness patterns, the so-called optical flow. Computing optical flow with an acceptable error is a difficult problem and has been studied extensively.However, most studies are limited to models with brightness constancy assumptions. In this paper, the authors study errors/biases in optical flow when a generic image brightness change model is used. An isotropic function, the spatial correlation function of the image brightness is used, to estimate errors/biases in optical flow. The results show that (1)errors always exist in optical flow when brightness changes, except for zero--motion; (2)the error is larger for more random images than that for smooth images ; and (3) the error can be quantitatively estimated based on the correlation function. In this paper, both theoretical derivation and real image experimental results are presented.