notes

手写识别是模式识别中研究得一个热点

1. later write a hysterysis "wrapper" to allow thresholding of feature
detectors over time (and to recognize start and stop of feature).

2. get our own unix group (dragons/sharks).

3. detectors output thier state (i.e. in zone, or not), and when state
changes event occurs.

4. This is the way this bizarre system works (nice and modular,
creating simple little modules): 

  a) Send the data file to each of the detectors individually, and
pipe the outputed state info to the event detector. Append the
outputed event info from each state/event pass to the same file.

  b) Sort the file by the time of the events (to order the events).
This info can then be passed to a classifier.

5.