Ask Your Question
0

Stereo Vision Distance Estimation Issue

asked 2017-05-16 08:39:13 -0500

Lexus09 gravatar image

updated 2017-05-16 09:12:24 -0500

Hi, I am attempting to calculate the distance to a given object in a pair of stereo images, using the SGBM algorithm. Depending on the calculated distance and my current coordinates, I am calculating the GPS coordinates of the given object.

Now the results are pretty average when I surpass the 2-meter mark and the results are very inaccurate. As the distance increases, the disparity map suffers greatly which inhibits the ability to accurately capture the object in the scene. Is there any way to get better distance results when testing on distances greater than the mentioned 2-meter mark? Or, do I consider this to be a limitation of my program?

Would highly appreciate any responses.

Thanks!

edit retag flag offensive close merge delete

1 answer

Sort by » oldest newest most voted
2

answered 2017-05-16 08:57:18 -0500

You can achieve higher accuracy if you compute the disparity at a higher resolution, but you will likely see a greater benefit from increasing the baseline.

edit flag offensive delete link more

Comments

Thank you for your reply. Do you think there is a certain depth limit on this matter? I am using 2 identical Logitech C270 webcams. Like do you think it could be viable for a distance of 20 meters for example?

Thanks!!

Lexus09 gravatar imageLexus09 ( 2017-05-16 09:02:42 -0500 )edit

Also, what I forgot to mention is that the disparity map becomes pretty average as distance is increased, hence the mediocre calculation

Lexus09 gravatar imageLexus09 ( 2017-05-16 09:05:35 -0500 )edit

Your error will increase quadratically with distance. Do the experiment of your finger in front of your face and see that when right in front of your nose a small change in the position of your finger results in a massive shift in the location of your finger in the image produced by each of your eyes fixated on infinity. Notice then how at arm's length, an equivalent change in depth produces tiny changes in the location of the finger in the image produced by each eye fixated on infinity. For cameras it's the same. It's the same reason that your ability to estimate distance to objects degrades as the objects recede and 3D scenes at a distance appear flat. You'll have to do the math of finding what your depth uncertainty is at a given distance given your image resolution and camera baseline.

Der Luftmensch gravatar imageDer Luftmensch ( 2017-05-17 08:13:37 -0500 )edit

I found the paper "Fast, Unconstrained Camera Motion Estimation from Stereo without Tracking and Robust Statistics" by Heiko Hirschmu ̈ller, Peter R. Innocent and Jon M. Garibaldi to be the simplest (and maybe good enough) method for computing stereo error. Look at section 2.3.2.

Der Luftmensch gravatar imageDer Luftmensch ( 2017-05-17 08:19:22 -0500 )edit
Login/Signup to Answer

Question Tools

1 follower

Stats

Asked: 2017-05-16 08:39:13 -0500

Seen: 281 times

Last updated: May 16 '17