This is a topic that seems to be of continued interest in this group.
Using a video camera is an approach I've not heard of before.  It seems
to be reasonably accurate and inexpensive compared to some of the other
schemes.  It is a bit rough around the edges but someone with a bit of
Linux programming skills should be able to get it working for just
about any format.  Take a look at

  http://www.ece.mtu.edu/~jarau/piano-scroll.html 

Bill Budenholzer

 [ I think the roll played is Supertone 5424 or Columbia 712, "Lonesome
 [ and Blue - Marimba Waltz", played by Nell Morrison, November 1923.
 [ -- Robbie

 [ I like the idea of using a HD 2D video camera as opposed to the 1D
 [ line-scan cameras that many of us have used in the past.   With just
 [ a simple distance calibration, the roll speed can be computed from
 [ observing successive scan frames, with the possible exception of
 [ very long silent periods.  An HD camera provides 1920 horizontal
 [ pixles.  This is almost 200 DPI across 10";  It should be enough
 [ resolution for even scanning very wide player-organ rolls.
 [
 [ Many of us like to preserve our original scans so that we can
 [ re-process the scans as we improve our detection algorithms.  I'm
 [ curious if anyone knows whether the MPEG video compression that is
 [ popular right now distorts the image of the moving holes in a way
 [ that's significant ?
 [
 [ Finally, Jacob (the project author), mentioned he had trouble
 [ detecting the holes in the area where the words were.  I think
 [ an appropriate amount of backlighting would make the holes stand
 [ out unambiguously.
 [
 [ Jacob has generously made his code available to the community.
 [ Since he's established how to capture the video and process it
 [ it would be a great starting place for those who want to try
 [ applying their own detection algorithms to captured video.
 [
 [ Jacob, Thank you!        --Jody