More on Optical Scanning
By Jody Kravitz
In response to Mark Fontana's mail above:
> CCD idea and was considering going this route except for the timing resolution
> problem. I never considered the vibration problem or the issue of measuring
> paper motion. But using a camera makes the system very versatile.
The "camera" is the solution, but also the problem. Because the "camera"
is coupled to the spoolbox by a lens it is possible to scan virtually any
size paper but _ANY_ motion in the "tripod" camera mount produces
time-varying errors in the scanned data. We've machined a bunch of little
jigs to accomodate differrent roll sizes and types of ends. Aeolean
residence organ rolls are a challenge because the tracker bar has two rows
of staggered holes -- the tachometer we designed was to allow us to delay
the second row of holes the right amount.
>
> My main objective was to design a system that would do the job in
> real time... because spending any longer per roll on a collection
> of hundreds (..thousands) of rolls would take way too long, and that's
> not even figuring in editing time and whatnot. I hope you find a
We were originally hoping to be able to scan up to 4x real time. We
also generate MIDI in real time (to a type 0 file and also to a MPU-401
so the operator can have some feedback), but we also keep the pixle data
for additional processing later. This makes it possible, after the fact,
to repair errors in the recording process (torn paper, wandering paper,
splotches on the paper, as well as improving algorithms for detecting
chain perforations).
Using a CCD camera generates a lot of "extra" data since not only do you
get data from over the centers of the holes, but lots of columns of
potentially irrelivant data in between. The extra data can be useful
for computing out artifacts. Thankfully cheap computes are getting really
fast. The data compresses down very nicely (we use PKWare's PKZIP on DOS,
GNU's GZIP on Unix) for data compression. Yes -- I do most of the
serious algorithm development on Unix and the port the results back to
DOS. We'll eventually have to bite the WINDOWS(tm) bullet, but I'm in
no hurry.
> suitable CCD device for your project-- have you thoroughly checked
> into camera units manufactured for industrial and robotics applications?
We think this would be a good idea, but we're not sure where
to look. None of my trade journals have been helpful. The one system
I investigated cost 10x what we're currently using! Further pointers
would be welcomed.
>
> I went the tracker-bar-fitted-with-phototransistors route. The photo-
> transistors are connected to the digital inputs of 4 8255-based digital I/O
> boards in a dedicated 386/40 PC. With this setup, each note on the tracker
> bar corresponds to a single port-mapped bit of memory. So to take a sample
> of the tracker bar, I simply read 12 bytes (88/8 = 11 + 1 byte for the pedal).
> The IR exciter source for the phototransistors is a 150 W flood lamp
> powered by a clean DC supply (with variable intensity to compensate
> for various opacities of paper in the rolls).
This is delightfully simple. We've not investigated, but I wouldn't
be surprised if this new paper is opaque at some (longer) wavelength.
Our scanner is not without its hassles. We have to deal with "spots"
on the paper. Word rolls must be scanned from the back side! We've
talked about the possibility of scanning the holes _AND_ the words
with a color CCD camera by doing some clever illumination. However
I fear we would have to go from 1 bit per pixle to at least 4 bits
per pixle per color. Even if we could get away with two colors,
this would increase the amount of data we have to get INTO the machine
by a factor of 8.
>
> The drive motor, exciter lamp, and transmission (play-reroll) are all
> controlled by the computer so that the process of scanning rolls is
> semi-automated. However, there is no auto-tracking system, so an
> operator must manually shift the spools during scanning to maintain
> proper tracking. I'd like to fix this someday...
This tracking problem is part of why we expect to have to reprocess
some of the pixle data. There are some pretty simple but elegant
pneumatic tracking (paper centering) systems around. Maybe one
of our other readers could recommend something.
Mark,
I want to compliment you on getting a project of this complexity
working while going to school.
Jody
P.S. Do you know anything about digital signal processing ? I've got
some great _OLD_ piano solo audio recordings I'd love to do some
notch filtering on...
|
(Message sent Wed 7 Jun 1995, 01:46:48 GMT, from time zone GMT-0700.) |
|
|
Mechanical
Music Digest™ Archives
