home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!munnari.oz.au!uniwa!uniwa!nfm
- From: udptech@uniwa.uwa.edu.au (Denis Brown)
- Newsgroups: comp.sys.transputer
- Subject: video digitising, neural nets
- Date: 11 Sep 1992 08:32:51 +0800
- Organization: The University of Westrn Australia
- Lines: 36
- Message-ID: <18opfjINNc1n@uniwa.uwa.edu.au>
- NNTP-Posting-Host: uniwa.uwa.edu.au
- X-Newsreader: Tin 1.1 PL5
-
- I have an interest in digitising stereoscopic video images for use in
- human movement studies (e.g. limb movement dynamics) and, unrelatedly
- at present, neural networks for various modelling tasks. From the
- data and experience to hand it seems that transputers offer the power
- and expansion prospects which will be required (especially in the
- neural nets). Accordingly I have budgeted for something like the
- Transtech TTG-F frame grabber and a 10-TRAM host card for my baseline
- environment (networked PCs). The frame grabber takes up 4 TRAM slots,
- so I expect to have a reasonable expansion path before having to get
- more host cards. I'm experienced in C, so have opted for this as the
- lingua-franca in the form of 3L's Parallel C.
-
- On the video (movement) side, I'd be interested in deltas (subtracting
- successive frames, effectively), with concentration on maybe 6% of the
- full scanned area. Any comments from anyone who's done/doing/given-up-
- on something along these lines ?? Granted, it's not the sort of soft-
- ware you'd knock up haphazardly before morning tea break one day, but
- according to one of our ex Prime Minister's, "Life wasn't meant to be
- easy!". BTW, I don't need display of real-time (or indeed ANY) video,
- just a bunch of numbers saying "this point moved xx,yy,zz units since
- the last capture time". Using multiplexed video sources into 1 grabber
- and CCIR TV specs, this means a timing resolution of 40 millisecs, more
- than fast enough for what I have in mind.
-
- Another way of going would be a workstation (SUN, DEC, etc) with a
- frame grabber card. There'd most likely be heaps of nearly-written
- code out there, but the comparative "isolation" of the processing horse-
- power from the grabber hardware weighs up in favour of the txputers, I
- think. Also I have a desire to implement the n/nets (later on) in a
- parallel environment.
-
- Comments, etc appreciated.
- Cheers,
- Denis
-
-
-