Stereo library 2012 pointers

[Klemen]

I understand, thanks!

So (it seems I cannot stop asking questions now) if I tilt both cameras by an equal angle, the intersection of both principal axes and cameras form an isoceles triangle. So the calculation is:

So, using this equation dZ=(dD*Z)/(2*f*sin(theta)) the tilted depth resolution can be calculated (provided, that the tilt of both cameras is the same)? So baseline has no weight here? Except to determine the minimum measuring distance of course.

The problem with this comercial webcameras is that in most cases no sensor size is specified. I only have diagonal FOV. I would need the actual sensor dimensions to calculate the vertical and horizontal FOV also.

Best regards,

K

[ronak25]

Hi Klemen,

    Nice to see your sustained interest. I really like your passion. Your analysis is mathematically attractive but slightly incorrect :). Actually, your analysis upto b =2Z*sin(theta) is correct. However, when you apply dZ = (Z*Z/(b*f))*dD equation, as this equation is a differential equation, you cannot directly equate b = 2Z sin(theta). Simply, this equation will not be valid as you have taken a differential over dZ. In this case, we actually apply the original equation and then rotate the dZ by theta. So, it will end up increasing. So, you basically cannot avoid dependency on baseline. 

   For second part of your post, yes there is a trouble getting the sensor size. However, if you are given a diagonal FOV, "assume" 3:4 aspect ratio and come up with the horizontal and vertical FOV. Now, if you know focal length of the camera and image resolution in pixels, you can approximately get the sensor size with 3:4 aspect ratio :)). This is crude but might be close to the original values. You may try your luck, if you feel this may be useful. I am 50-50 about this.

Regards,

Ronak.

[Klemen]

Hi Ronak,

so basically I take the general equation dZ = (dD*Z^2)/(f*b) and multiply the result by cosine theta, where theta is the tilt angle of the cameras? If cameras are paraller, then this angle is 0° (cosin 0°=1), thus the equation reverts back to the general one.

I will try to calculate the sensor size and see what I come up with. I already know the answer though - web cameras are not good for accurate stereo vision 🙂

Best regards,

K

[ronak25]

Hi Klemen,

   only one correction :). Divide the depth resolution by cos(theta), rather than multiplying by it.

Regards,

Ronak.

[Klemen]

Thank you for the prompt answer.

But still somethin is unclear to me. If I use the equation dZ = (dD*Z^2)/(f*B) with the same parameters and only increase the focal length than dZ would be smaller, thus better depth resolution right? On the other hand if I divide the dept resolution by cosinus theta it gets worse by increasing the angle theta (the dZ increases) . So, the best depth resolution is when using the parallel setup, but then the baseline has to be small to cover the measuring area?

Basically you have to achieve a good trade-off between all the parameters (wide baseline and small tilt) depending on what you want to measure?

Best regards,

K

[ronak25]

Exactly. There is no 'golden' stero set-up. there is always a tradeoff depending on what you want to measure with what accuracy. There is no rocket science.

Regards,

Ronak.

[34thsecondstreet]

hello 

I read all of the posts on the topic(solving problem for stereo diffrent camera)...

but I want to know if I use two same DSLR cameras  with high resolution (5400*3400pixels) ,the Labview programming can't help us.

I say this base on your statments for depth map that Mr.ronak said 'use photos with resolutions 600*400 '.

I say this because I need more accuracy.

special thanks tor your posts ,It's help me so much

I will be pleased if you can help me...

with best regards