Sabrina, answers below...kk

On Tue, Feb 8, 2011 at 6:07 AM, Sabrina Kliegl <skliegl@zitmail.upb.de> wrote:
> Hi Kurt,
>
> thank you, that is what I was looking for. Maybe you can answer me some
> questions about the current OpenCV implementation of the block matching:
>
> Did I get it right, that you are using a horizontal Sobel filter now
> instead of the LoG (as mentioned in the paper)? If so, why only a
> horizontal Sobel filter and no vertical Sobel as well?

X-Sobel gets rid of subpixel error amplification.  See this paper:

Stereo Matching in the Presence of Sub-Pixel Calibration Errors
Heiko Hirschmu¨ ller, Stefan Gehrig

>
> Is the block matching applied to the images resulting from the edge
> detector or to the sharpend original images?
>

Block matching is applied to the X-Sobel images.

> And I wonder how you achieve subpixel accuracy. I suppose the last
> magical lines of findStereoCorrespondenceBM(...) to do this:
>
> int p = sad[mind+1], n = sad[mind-1], d = p + n - 2*sad[mind] +
> std::abs(p - n);
> dptr[y*dstep] = (short)(((ndisp - mind - 1 + mindisp)*256 + (d != 0 ?
> (p-n)*256/d : 0) + 15) >> 4);
>
> But I don't get the idea behind it. Can you give me a hint?
>

Subpixel interpolation is done by a quadratic interpolation on the min
correlation value and its two neighbors.  If you work it out, the
fractional disparity increment is:

(p-n)/(2*(p+n-2*c))

where c is the min correlation value, p is the one preceeding it, and
n is the one after it.  The OpenCV version mangles this a bit to get
it into fixed-point form.

> Regards,
> Sabrina
>
> Am 08.02.2011 14:49, schrieb Kurt Konolige:
>> Sabrina, this is still the best publication I have for this:
>>
>> K. Konolige
>> "Small Vision Systems: hardware and implementation"
>> Eighth International Symposium on Robotics Research
>> Hayama, Japan, October 1997
>>
>> Cheers --Kurt
>>
>> On Tue, Feb 8, 2011 at 1:40 AM, Sabrina Kliegl<skliegl@zitmail.upb.de>  wrote:
>>> Hi everybody,
>>>
>>> I'm using stereo_image_proc in my diploma thesis and I wonder if there
>>> is any publication of Kurt Konolige (or somebody else) which explains
>>> the pre- and postfiltering and which I could use as a reference. Thanks!
>>>
>>> Regards,
>>> Sabrina
>>> _______________________________________________
>>> ros-users mailing list
>>> ros-users@code.ros.org
>>> https://code.ros.org/mailman/listinfo/ros-users
>>>
>> _______________________________________________
>> ros-users mailing list
>> ros-users@code.ros.org
>> https://code.ros.org/mailman/listinfo/ros-users
>>
> _______________________________________________
> ros-users mailing list
> ros-users@code.ros.org
> https://code.ros.org/mailman/listinfo/ros-users
>