简体繁体 English

从两个2D立体图像生成3D（x，y，z）点云

[英]Generating 3D (x,y,z) Point Clouds from Two 2D Stereo Images

原文 2014-06-01 02:32:21 2 1 python/ c++/ matlab/ image-processing

I've been trying to read through the Stackoverflow questions for generating point clouds (x,y,z) coordinates from a left and right stereo image pair. 我一直在尝试阅读Stackoverflow问题，以从左右一对立体图像对生成点云（x，y，z）坐标。

I haven't come to any definite solution, and I'm asking the community here for some help. 我还没有确定的解决方案，我想向这里的社区寻求帮助。

Problem statement: Given two stereo images, generate 3D (x,y,z) cartesian coordinate point clouds from those and do so in a way that lends itself to completing this point-cloud generation in a way that could work over a large set (thousands) of pairs of stereo images 问题陈述： 给定两个立体图像，从中生成3D（x，y，z）笛卡尔坐标点云，并以一种可以在大型集合上工作的方式完成此点云生成的方式（成对的立体图像

My programming language experience lends itself to MATLAB, but I've dabbled in Python, and C++ is limited, but I may be able to work in that as well. 我的编程语言经验很适合MATLAB，但是我已经涉足Python，C ++受到限制，但是我也可以在其中工作。

Speed is a factor here, so the the idea is to find a quick method of successively going through these pairs and generating the point cloud. 速度是此处的一个因素，因此其思想是找到一种快速方法来依次遍历这些对并生成点云。

NOTE: I am not asking for the BEST as to avoid comparative solutions, I'm just asking for solutions. 注意：我并不是在寻求最好的解决方案，因为它避免了比较的解决方案，我只是在寻求解决方案。

Thank you very much! 非常感谢你！

Edit: After being recommended to utilize the Stereo Block Matching implementation in OpenCV, I may steer away from this method, as there is a wide variation in regards to texture of the topographical pictures that I'm dealing with. 编辑：被推荐在OpenCV中使用“立体块匹配”实现后，我可能会避开此方法，因为我要处理的地形图片的纹理存在很大差异。

1 个解决方案

What you probably want is called "Stereo Block Matching". 您可能想要的就是“立体声块匹配”。 If you don't feel like writing it yourself, OpenCV has an implementation ready to go (even a CUDA accelerated version ). 如果您不想自己编写它，则OpenCV提供了一个可用的实现（甚至是CUDA加速版）。