New demonstration regarding stereo vision

[juliena82]

@applemuncy

Following the idea you have submitted in the "demonstration proposal" thread, I propose that we try to define together more precisely the perimeter of this demo you are thinking about.

As a beginning:

• This demo should open 2 cameras (needs some update in the common engine)
• Would we include stereo camera calibration ? Or maybe we can make a separate demo for stereo calibration, and this demo would only retrieve the results of previous calibration ?
• We could also include in the stereo calibration demo the computing of the 2 rectification maps
• Then the stereo vision demo would compute (in real time) the disparity map, and optionnaly the depth (e.g. Z = f(X,Y))

So we may have at least 2 démos to specify : the stereo calibration, and the rectification demo.

And from that point surely we can do many more things with reprojected video:

• incrustation of virtual 3d objects
• localization in space (e.g. X,Y,Z) of a specific object, detected through interest points (for instance, it could be very usefull to track precisely the position of a robot / ball / something in real time)
• any other ideas welcome...

What do you think about it? I am sure you have other ideas, please feel free to say! Also, maybe you have already written some code about thèse subjects?

[applemuncy]

I think you have captured most of my thoughts and added new ideas. Toward this goal I have purchased 3 pairs of very inexpensive webcams that I I still need to mount side by side mindful of the future needs of physical adjustments. The only code I have done so far is minimal to get working two cams streaming at the same time. I do see the future need of a specialized demo program.

Perhaps a hypothesis to test: Is there much to be gained by using two cameras side by side rather than a single camera that perhaps moves? Anyway, coding is good : )

[juliena82]

I think that a single camera that moves might need quite other algorithms than 2 fixed cameras. Also I don't know if there is any OpenCV function to handle this case...

[valera-rozuvan]

@juliena82 @applemuncy I think we can solve the problem of 2 cameras if we provide the user with ability to connect IP cameras. Please see Issue https://github.com/tsdconseil/opencv-demonstrator/issues/38 .

[applemuncy]

Two webcams do work now provided underlining hardware is fast enough, eg. USB.

[applemuncy]

I an testing using addWeighted() to combine the two images into one as a means of aiming the two webcams at a distant point by physical adjustment.

[juliena82]

Good idea! Don't hesitate to add it to the main repo when it works.