Re: [Paparazzi-devel] Photogrammetry (was georeferencing video stream)

[Steve Joyce]

You know for relatively flat areas is might not be so bad at all if you are 
careful to synchronize your attitude data with the video frames.

What I would do is start with trying to project video frames on to a flat 
plane at ground level.  There is no need to mix in GRASS or other software, 
everything you need is in the OpenCV library (although it might not be 
obvious at first), including processing video sequences.  See 
cvWarpPerspective().

I think the biggest error you will have will be in yaw.  In the abscence of 
a compass, you can only infer the yaw angle from the GPS heading and the 
current wind estimate. You can try to extract some keypoints from the video 
frames and match keypoints between frames to adjust the angles.  I think 
there is a demo progam in OpenCV that tracks keypoints in video sequences 
(blobtrack?).

You were pretty close with your plan you wrote a few days ago except you 
don't actually have to rectify the entire image until you want to produce a 
final result.  All the calcualtions are done with the orientation data and 
tie point cloud.
[*A] is a projection from image coordinates to object coordinates.  See 
cvProjectPoints2(), then realize it is backwards, but the equations are well 
known.
[*B] is essentially a block adjustment which I tried to explain yesterday.

/Steve
----- Original Message ----- 
From: "Chris Gough" <address@hidden>
To: <address@hidden>
Sent: Friday, October 09, 2009 12:26 AM
Subject: Re: [Paparazzi-devel] Photogrammetry (was georeferencing video 
stream)


That's an awsome response, thank you.

Sounds like the "direct georeferenced video stream" approach is
fesable, but likely to give me dissapointing results (be of dubious
value). I think I'll try to learn patience and more about block
adjustment and the other things you have linked to.

Chris