Re: cvCalcOpticalFlowPyrLK error-value calculation.

Hi,

I also want to use the error values returned by the OpenCV function 
cvCalcOpticalFlowPyrLK, but I don't know how to normalize the values with the 
intersection 
size. I would be really glad, if somebody have some details, how to use these 
error values. 
And what is the intersection size described in the previous message and which 
values I 
have to use to calculate it. Or which values I have to use for this equation: 
(maxJ.y-minJ.y)*(maxJ.x-minJ.x).

Stefanie






--- In OpenCV@xxxxxxxxxxxxxxx, "Kibkalo, Alexander" <alexander.kibkalo@xxxx> 
wrote:
>
> In current implementation the error value is summed on the intersection
> of the window and the image. For points far from the border this
> intersection coincides with the window, for points near the border is
> less than it. You can normalize it with the intersection size
> (maxJ.y-minJ.y)*(maxJ.x-minJ.x)
> 
> Alexander 
> 
> -----Original Message-----
> From: OpenCV@xxxxxxxxxxxxxxx [mailto:OpenCV@xxxxxxxxxxxxxxx] On Behalf
> Of andreasbernatz
> Sent: Tuesday, November 29, 2005 5:46 AM
> To: OpenCV@xxxxxxxxxxxxxxx
> Subject: [OpenCV] cvCalcOpticalFlowPyrLK error-value calculation.
> 
> Hi out there,
> 
> I'm using the cvCalcOpticalFlowPyrLK function for a long time already,
> to track some features in an image in order to calculate the egomotion
> of my cam. To enhance this calculation I thought of using 'weights'
> attached to every feature, telling me, how much I can trust the last
> retrived movement. I thought it might be a good idea to use the values
> of the "track_error" array to calculate these 'weights'. So here comes
> my question: What exactly tells me a "track_error"-value, how is it
> calculated? and what are the bounds for the possible values (cause I
> like to normalize them to values whithin [0..1]).
> 
> I tried to slove this question with help of the sourcecode, but I'm
> not sure I understood everything right inside.
> (eg: to me its looking like the errorvalue is depending on the size of
> the 'patch' somehow, but on the same time the size of the patch is
> lookin variable dependin on the position of the feature within two
> images(eg. at the border it might be smaller ??), so how can the
> errorvalues be compareable??)
> 
> Waiting hopefully for your enlighting answers,
> 
> Andreas
> 
> 
> 
> 
> 
> 
> 
> Change settings: http://www.yahoogroups.com/mygroups, select
>    Get Emails (get all posts)
>    Daily Digest (one summary email per day)
>    Read on the web (read posts on the web only)Or Unsubscribe by mailing
> OpenCV-unsubscribe@xxxxxxxxxxxxxxx
>  
> Yahoo! Groups Links
>





Change settings: http://www.yahoogroups.com/mygroups, select
   Get Emails (get all posts)
   Daily Digest (one summary email per day)
   Read on the web (read posts on the web only)Or Unsubscribe by mailing 
OpenCV-unsubscribe@xxxxxxxxxxxxxxx