针对双目立体匹配中左、右图像存在平移、缩放、旋转问题,提出了一种新的基于KAZE特征的双目立体匹配方法。通过AOS算法和可变传导扩散方法来构造非线性尺度空间,然后检测感兴趣特征点,计算特征点的主方向,并且基于一阶微分图像提取具有尺度和旋转不变性的特征描述向量。该KAZE特征描述向量具有平移、旋转不变性和尺度不变性,也具有很强鲁棒性。双目立体图像匹配实验结果表明,即使在旋转、缩放、角度变换等情况下,该算法依然保持很高的匹配率。
This paper presents a new based on Kaze characteristics of binocular stereo matching method.First,through the AOS algorithm and variable conduction diffusion method to construct nonlinear scale space,then detected features of interest points,calculation of the main direction of the feature point,and based on the first-order differential image extraction with scale and rotation invariant feature vector description.