GestureLab |
Monash
University
Clayton,
Victoria, Australia |
What
is GestureLab?
GestureLab
is a software tool designed to facilitate rapid development and testing
of domain specific gesture recognizers. Recognizers can be developed
entirely
within GestureLab without any need for a testbed application and can be
coupled with grammar engines such as Cider without modification.
GestureLab recognizers follow the standard approach to statistical
gesture recognition: recognition is performed on digital ink which
includes position, timing,
pressure, and angle data. Statistical summary features such as
the total length of the gesture, initial stroke angle, and maximum
curvature are extracted from this
data and used by a classifier algorithm to predict class labels
(gesture types). A recognizer thus consists of a bundle of feature
extractors and a classifier
algorithm trained on a particular gesture corpus.
GestureLab supports all phases of the recognizer development process:
(a) collecting, manipulating and sharing gesture corpora, and (b)
automatic training and
cross-validation of feature extraction and recognizer mechanisms.
In the event that the built-in feature extraction and recognizer
mechanisms are insufficient,
GestureLab also allows the developer to readily define (c) new feature
extraction mechanisms and (d) new recognizer algorithms. GestureLab
generates probabilistic
recognizers that return membership probabilities for all possible token
classes instead of a single most likely class. Probabilistic
classificators can be useful
during disambiguation at a syntactic level.
Version
1.0-alpha4
- GestureLab-1.0a4.zip
(18.8Mb) contains the GestureLab application for Mac OS X (10.4
onwards) and a default database of gestures.
- digest.zip (15.8Mb)
contains all GestureLab-1.0a4 source code.
- librecognizer.zip
(600kb) contains the source code for librecognizer, a library to
develop and integrate the default GestureLab plugins.
- Documentation
(627Kb) for this version.
|
|