depth estimation using stereo cameras

Hi, I want to implement depth esimation with stereo cameras.

after rectification, the images should follow this rule: objects that are closer to the cameras should have large disparity and vice versa.

but in my results there is a far object with relative big disparity. I have noticed that in the following image, that I took from matlab example, there is a similar case (the car at the end of the parking lot) .

My question is why does this happen? and would it damage the depth estimation?

I'll just add that the I used the calibration app one time, and feature detector and estimateFundamentalMatrix function a second time. both gave similar results.

Answer 1

[Dima Lisin]

Hi Idan,

When you do uncalibrated rectification, the rectified views do not always end up parallel, but may be verged a little. That means the disparity may actually be negative for objects that are far away. I think this is what happened to the car at the end of the parking lot. In a case like this, if you compute the disparity map, you should subtract the minimum disparity value from the map. This will shift the range, to make the minimum disparity 0.

[Idan Dekel]

Thanks you Dima! I have a related question: Is it possible to get negative disparity very close to the cameras? It looks like I'm getting large negative disparity very near and very far from the cameras.

Thanks again

[Dima Lisin]

This can happen if camera 1 is to the right of camera 2, as you look in the same direction as the cameras. In that case, the sign of disparity values is reversed. Generally it is a good idea to make sure that camera 1 is on the left.