噪声环境下光流场估计方法

摘要

Noise disturbance is a key problem which should be taken into account in accurate optical flow estimation, since image data may be blurred in capture and transmission process. To resolve this problem, firstly, the two hypotheses of traditional optical flow method were discussed, through which the limitations of the luminance constant hypothesis were unveiled; and then, according to the noise.constraint, the basic optical flow equation was corrected based on the MUKAWA' s results; after that, the global smoothness constraint of optical flow field was introduced to regularize the problem of optical flow estimation, based on the new basic optical flow equation; finally, the arithmetic expressions of determining optical flow field were deduced from the new basic optical flow equation and the global smoothness constraint using the variational approach. The CAVIAR video data were used to test our method. Firstly, the test data werefedded in the Gaussian noise with different levels; and then, the optical flow fields of these blurred data were determined using traditional methods and our method, respectively; finally, the experiment results with different methods were analyzed and compared. Both of the test results and comparative results showed that our method outperforms the others in the presence of noise.%图像在采集、传输过程中可能会受到噪声污染,噪声干扰是准确进行光流场估计必须考虑的技术难题.为解决这一难题,首先,对传统光流法的两条基本假设进行了讨论,指出了灰度恒定假设的局限性；然后,在MUKAWA推论的基础上,根据噪声约束对基于灰度恒定假设的光流基本方程进行了修正；而后,在新的基本方程基础上,引入全局平滑约束使光流估计问题正则化；最后,采用变分方法推导了噪声环境下光流场的估计算式.为测试算法性能,对CAVIAR视频数据加入不同水平的高斯噪声,而后分别采用传统光流方法和本文方法进行鲁棒性实验,最后对实验结果进行了分析比较.实验和分析结果证明本文方法具有更好的抗噪性能.