序列图像运动自适应V1—MT光流估计算法

发布时间：2018-05-28 03:07

本文选题：光流 + V　；参考：《工程科学学报》2017年08期

【摘要】：针对传统算法在抗光照变化影响、大位移光流和异质点滤除等方面的不足,从人类视觉认知机理出发,提出了一种基于机器学习和生物模型的运动自适应V1-MT(motion-adaptive V1-MT,MAV1MT)序列图像光流估计算法.首先,引入基于ROF模型的结构纹理分解(structure-texture decomposition,STD)技术,有效解决了光照和色彩变化的影响.其次,利用多V1细胞加权组合及非线性正则化模拟MT细胞模型,并结合岭回归训练学习得到运动自适应的权重,解决对目标的运动速度感知问题.最后,引入由粗到精的增强方法和图像金字塔局部运动估计采样,将V1-MT运动估计模型应用于实际大位移视频序列.理论分析和实验结果表明,新方法能更加拟合人眼视觉信息处理特性,对视频序列具有普适、有效、鲁棒的运动感知性能.
[Abstract]:In view of the shortcomings of traditional algorithms in the aspects of anti-illumination effect, large-displacement light flow and heterogeneous point filtering, the mechanism of human visual cognition is considered. An optical flow estimation algorithm for moving adaptive V1-MT(motion-adaptive V1-MTMAV1MT-1 images based on machine learning and biological model is proposed. Firstly, the structure texture decomposition (STD) technology based on ROF model is introduced, which effectively solves the influence of illumination and color changes. Secondly, the multi-V1 cell weighted combination and nonlinear regularization are used to simulate the MT cell model, and combining with the ridge regression training, the weight of motion adaptation is obtained to solve the problem of target velocity perception. Finally, the coarse to fine enhancement method and the local motion estimation sampling of image pyramid are introduced, and the V1-MT motion estimation model is applied to the actual video sequences with large displacement. Theoretical analysis and experimental results show that the new method can better fit the characteristics of human visual information processing, and has universal, effective and robust motion perception performance for video sequences.
【作者单位】：空军工程大学航空航天工程学院;陕西服装工程学院;
【基金】：国家自然科学基金资助项目(61372167,61379104)
【分类号】：TP391.41

【相似文献】