如何视野改变立体视觉中的深度估计？ [英] How Field of view changes depth estimation in stereo vision?

问题描述

我想用两个摄像机估计立体声系统的深度。我使用的简单方程是：

I'm trying to estimate depth from a stereo system with two cameras. The simple equation that I use is:

           Baseline*Focal
Depth = ----------------------
             Disparity

两个摄像机的视图不会改变允许的最大深度？
它只改变最小深度可测量？

The field of view of the two cameras doesn't change the maximum depth allowed? It changes only the minimum depth measurable?

推荐答案

在顶端，可测量的深度受限于所使用的摄像头的分辨率，这反映在视差中。随着深度变得更大，视差倾向于为零。利用更大的视场，在较低深度处它将有效地为零。因此，更大的视野可以降低最大深度的可测量，但你可以通过使用更高分辨率的相机来补偿一些。

At the top end the measurable depth is limited by the resolution of the cameras you use, which is reflected in the disparity. As depth becomes greater disparity tends to zero. With a greater field of view it will effectively be zero at a lower depth. Thus a greater field of view lowers the maximum depth measurable, but you can compensate somewhat by using higher resolution cameras.

澄清：你应该注意到正确）你测量像差的像差，但然后将其转换为米（或毫米，如下所示）。那么完整的公式是：

To clarify: you should note that (if you do things correctly) you measure disparity in pixels but then convert it to meters (or milimeters as I do below). The full formula is then:

          Baseline * Focal length
Depth = ----------------------------
        Pixel disparity * Pixel size

假设您有以下设置：

Baseline (b) = 8 cm (80 mm)
Focal length (f) = 6.3 mm
Pixel size (p) = 14 um (0.014 mm)

可以测量的最小差异为1像素。使用已知数字，这转换为：

The smallest disparity you can measure is 1 pixel. With the known numbers this translates to:

Depth = (80*6.3)/(1*0.014) = 36,000 mm = 36 m

在这种情况下，这将是你的上限。请注意，在此范围内，您的测量值非常不准确。下一个可能的差异（2像素）发生在18米的深度，接下来（3像素）在12米等。将基线加倍将范围加倍到72米。加倍你的焦距也会使你的范围加倍，但注意两者都会在短期对你产生负面影响。

So in these circumstances this would be your cap. Note that your measurement is wildly inaccurate at this range. The next possible disparity (2 pixels) occurs at a depth of 18m, the next after that (3 pixels) at 12m, etc. Doubling your baseline would double the range to 72m. Doubling your focal length would also double your range, but note that both would negatively affect you at the short end. You could also increase your maximum depth by decreasing the pixel size.

在像素大小为0.014 mm的情况下，您可能会说的CCD的水平分辨率为1024像素，对于大约14.3mm宽的CCD。如果您将同一区域中的像素数量加倍，您可以将最大范围加倍，而不会在近端丢失任何东西（因为限制由基线和焦距确定，保持不变）。

At a pixel size of 0.014 mm, you are probably talking about a CCD with a horizontal resolution of something like 1024 pixels, for a CCD of about 14.3 mm wide. If you double the number of pixels in the same area you would double your maximum range without loosing anything at the near end (because the limitations there are determined by baseline and focal length, which stay the same).

这是一个非常好的概述，在深度测量中的权衡立体视觉。
和wikipedia的这篇文章有一些很好的信息关于像素大小，ccd大小，焦距和视野。

This is a very good overview of the tradeoffs in depth measurement in stereo vision. And this article on wikipedia has some good info on the relationship between pixel size, ccd size, focal length and field of view.

这篇关于如何视野改变立体视觉中的深度估计？的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！