新的单目立体视觉的视差图的获得方法

在立体视觉中，视差间接反映物体的深度信息，视差计算是深度计算的基础。常见的视差计算方法研究都是面向双目立体视觉，而双焦单目立体视觉的视差分布不同于双目视差，具有沿极线辐射的特点。针对双焦单目立体视觉的特点，提出了一种单目立体视差的计算方法。对于计算到的初步视差图，把视差点分类为匹配计算点和误匹配点。通过均值偏移向量（Mean Shift）算法，实现了对误匹配点依赖于匹配点和图像分割的视差估计，最终得到致密准确的视差图。实验证明，这种方法可以通过双焦立体图像对高效地获得场景的视差图。

Extracting disparity map from bifocal monocular stereo vision in a novel way

Disparity is a key point in stereo vision as it shows the depth information of the scene indirectly, and disparity calculation is the basis of the depth calculation. The traditional disparity calculation methods are all targeted at binocular stereo. However, compared with disparity in binocular, the disparity in monocular stereo is radial along epipolar line. Concerning the characteristics of bi-focal monocular stereo vision, an approach to get disparity map from bi-focal images was proposed in this paper. Preliminary depth map was obtained by matching cost calculated. By using Mean-Shift algorithm, discontinuities and misunderstandings in primary depth map were smoothed and eliminated according to the matched points and graph cut result. The experimental results show that this method can get disparity map efficiently from pairs of bi-focal images.