基于小波滤波器组的光流估计

Wavelet Filter Bank Based Optical Flow Estimation

为了解决光流估计时存在的孔径问题和时域混叠 ,在 Bernard的理论基础上 ,给出了基于快速滤波器组的方法 :小波函数在频域可以表示为滤波器组的乘积 ,所以时域的小波系数由傅里叶变换转换到频域 ,通过滤波器组快速求解。由于解析小波具有实小波所没有的优点 ,采用解析小波计算光流。文中给出该方法的结果 ,还给出了 Horn-Schunck的正则化方法的结果。比较后可知 ,本文的方法能较好地对运动物体进行光流估计。

To solve two problems existing in optical flow estimation i.e. aperture problem and time aliasing problem, on the basis of Bernard theory, a method based on fast filter bank is presented: Due to the fact that wavelet function can be formulated as the product of filter bank, wavelet coefficients in time domain can be transformed into that in frequency domain. In this way, the problem through filter bank is fleetly solved. With consideration of advantages of analytical wavelet over real wavelet, the analytical wavelet is used. Results of using the method and Horn-Schunck method are given. The method is superior to Horn-Schunck method in optical flow estimation.