Image format for Stereo Vision.vi

[gbvbalint]

Hello!

I would like to try LabVIEW's built in example for Stereo Vision. As far as I tried about two days, I couldn't get the depth image from my pictures. I have tried many image formats and sizes, but none of them worked (the depth map was full black). Is there any trick with the NI pictures, or the example only works with them? Or I should change some parameters in the program? I would appreciate any help because I don't have any more idea why doesn't it work.

Thank you!

gbbalint

[MoviJOHN]

First you need two cameras, placed ~2 inches apart.  Ideally, the horizontal and vertical axes should converge on some point in the field of view.

Once you have the two images, you should be able to get your depth map.  Note that homogenous surfaces will not show depth (all black) since there are no features to lock on to.  If you want to extract depth data in these areas, you need to project a scatter pattern.

[gbvbalint]

It's okay, I already have the pictures, but they don't work with the example. And as far as I tried, I couldn't get the depth map from any picture (I have downloaded from here too, from another posts, none of the worked) except those shipped in the example. So I'm curious what could be the problem.

[MoviJOHN]

Post your pictures.  I'll take a look.

[gbvbalint]

Here you are. I will attach it in two parts.

[gbvbalint]

the second part

[MoviJOHN]

I'm looking at the images now.  The first thing that strikes me is that the images are too dark.  I'll need to look in detail to determine if this is the case, but for dot grid calibration, you need good contrast between dots and the background.

The second thing, and this is just something I think I recall, is that the subsequent grid images need to cover the entire field of view of both cameras.

Ex.  For the first calibration frame, the grid needs to be in full view, meaning that you need to see all four corners.  In the subsequent views, the grid needs to cover the image fully, meaning there can be nothing else in the image, other than dots from the grid.

I'm going to mormalize your iamges, and see if results improve, and get back to you on this.

[MoviJOHN]

I ran your pictures through the Stereo Vision Example, and they are fine.  The only thing I was not immediately able to do was set good parameters for the disparity, so the results looked a little starnge, but I did get results.

I did have to hardwire the path to the image location, otherwise the VI expects the images to be in a folder relative to the VI path.

[lizhuo_lin]

Hello! MoviJOHN

do you get good depth image? I only got this.

[gbvbalint]

Hello!

I have found out something. If I set disparity about 64-80,corresponding windows size to 5, and depth range must be maximum about 90, i got the depth image, form these pictures. Then I just dont know how to set properly the parameters not with just trying.