摘要
针对计算机视觉中的运动估计问题,提出了新的基于纹理流的图像序列稠密运动估计算法,与常规基于恒定亮度假设的光流估计方法相比具有更好的稳健性、更高的准确性和更快的计算速度.在假设相对稳健的图像纹理特征不变的基础上,推导得到运动估计方程及运动向量.通过定义纹理特征的差异度,综合矩阵行列式值以及条件数来判断运动估计的准确性和稳定性,并以常用的Gabor滤波器组为例设计及实现了方法.通过使用多类视频序列、与各种典型方法得到的结果对比表明,该方法对复杂背景和运动目标具有独特纹理特征的情况下具有更好的估计效果.
Motion estimation is one of the basic problems in computer vision.A novel method based on textural flow was proposed in order to allow estimation of dense motion from image sequences.Compared with the conventional optical flow method using the assumption of constant brightness,it offers better robustness,higher accuracy,and faster computational speed.Based on robust invariant image textural features,a texture motion equation and motion vectors were derived.Difference levels,matrix determinant and condition numbers were defined and then integrated to judge the accuracy and robustness of motion estimation.In implementing the algorithm,a classic Gabor filter band was used.The algorithm was tested with many different kinds of image sequences and the results compared with those of typical methods.The results showed that the proposed method is best used when the moving targets have unique textural features or the background is complex.
出处
《哈尔滨工程大学学报》
EI
CAS
CSCD
北大核心
2010年第4期438-443,共6页
Journal of Harbin Engineering University
基金
国防基础研究基金资助项目(2004AA742209)
