4

Im a newbie to OpenCV, Ive built openCV and installed Qt to run the example http://code.google.com/p/opencvstereovision/source/checkout with success. The result is left and right webcam captures and after calibration a rectified image for both and a depth map. I want to do the following. I will click on an object to focus in any of these left or right images. Then I want to detect the corresponding pixel in the other image and calculate the distance to this object. I dont know what to do now and I need guidence. Maybe if I first calculate the corresponding point, I may work the rest out. How can I do this?

dramaticlook
  • 653
  • 1
  • 12
  • 39

2 Answers2

4

It would seem you have a good idea of what's required. This is a bit of a can of worms - so as ever I would recommend getting hold of the famous Hartley & Zisserman book for the canonical explanation. Here's a link to the relevant chapter.

But in brief...

I've not used the opencvstereovision wrapper class directly, but it sounds like it has taken the headache out of calibrating (camera intrinsics and extrinsics) and has calculated rectification via the Homography matrix (H) for planar geometry, or Fundamental matrix (F) for more complex epipolar geometry.

Probably similar to this original post.

What this rectification then means is that it has established the mathematical mapping between the same point in each image.

In a previous answer (from me) you can then do the maths by using the Fundamental matrix to perform triangulation - i.e. calculate distance.

However, note that this distance is only in the image coordinate frame (i.e. in pixels).

What is actually required to perform "real world" measurements (i.e. actual physical distance) is the calculation of the Essential matrix (E) which combines the Fundamental matrix and the cameras intrinsics (K) to, if you will, project the distances into the real world.

Community
  • 1
  • 1
timlukins
  • 2,694
  • 21
  • 35
  • Thanks man. The link I mentioned was the exact implementation from Chapter 12 and was explained in detail in the book. I guess cvComputeCorrespondEpilines, which was mentioned in page 427 also, is what is needed right? I will pass the pixels in one image and It shoud find the corresponding on the other image??? – dramaticlook Jan 19 '13 at 00:33
1
class StereoVar
{
StereoVar();
StereoVar(    int levels, double pyrScale,
                                int nIt, int minDisp, int maxDisp,
                                int poly_n, double poly_sigma, float fi,
                                float lambda, int penalization, int cycle,
                                int flags);
virtual ~StereoVar();

virtual void operator()(InputArray left, InputArray right, OutputArray disp);

int        levels;
double    pyrScale;
int        nIt;
int        minDisp;
int        maxDisp;
int        poly_n;
double    poly_sigma;
float    fi;
float    lambda;
int        penalization;
int        cycle;
int        flags;

...
};

Refer to: http://docs.opencv.org/modules/contrib/doc/stereo.html

Abhishek Thakur
  • 16,337
  • 15
  • 66
  • 97
  • can you please share any example code to make disparty map. I saw the stereo_match.cpp sample code.. but there is no explation, so it is really heard to understand – MMH Dec 03 '13 at 04:24