The Gesture RElative Accuracy Toolkit (GREAT) is an application and associated reusable library (DLL)
named RelativeAccuracyMeasures.dll for calculating relative accuracy measures for a set of gestures.
The 12 accuracy measures capture what happens during stroke gesture articulation, and are therefore more
revealing than overall recognition accuracy. For example, one measure is Shape Variability (ShV), which
captures how much a gesture's shape deviates along its path from a reference gesture. Reference gestures
are determined by computing the average gesture from a set, and then finding which articulated gesture is
closest to that average. The library provided encapsulates the computing of all 12 measures and can be
reused in other applications.
The 12 relative accuracy measures defined in the library are:
Shape Error (ShE): Average spatial deviation from a reference gesture.
Shape Variability (ShV): Total spatial deviation from a reference gesture.
Length Error (LE): Amount of "stretch" relative to a reference gesture.
Size Error (SzE): Amount of space consumed relative to a reference gesture.
Bending Error (BE): Average "turn" relative to a reference gesture.
Bending Variability (BV): Total "turn" relative to a reference gesture.
Time Error (TE): Average temporal deviation from a reference gesture.
Time Variability (TV): Total temporal deviation from a reference gesture.
Speed Error (VE): Average deviation in speed compared to a reference gesture.
Speed Variability (VV): Total deviation in speed compared to a reference gesture.
Stroke Count Error (SkE): Difference in number of strokes compared to a reference gesture.
Stroke Ordering Error (SkOE): How similar the stroke ordering is compared to a reference gesture.
Video
Our Gesture Software Projects
$Q: Super-quick multistroke recognizer - optimized for low-power mobiles and wearables
$P+: Point-cloud multistroke recognizer - optimized for people with low vision
$P: Point-cloud multistroke recognizer - for recognizing multistroke gestures as point-clouds
$N: Multistroke recognizer - for recognizing simple multistroke gestures
$1: Unistroke recognizer - for recognizing unistroke gestures
AGATe: AGreement Analysis Toolkit - for calculating agreement in gesture-elicitation studies
GHoST: Gesture HeatmapS Toolkit - for visualizing variation in gesture articulation
GREAT: Gesture RElative Accuracy Toolkit - for measuring variation in gesture articulation
GECKo: GEsture Clustering toolKit - for clustering gestures and calculating agreement
Vatavu, R.-D., Anthony, L. and Wobbrock, J.O. (2013).
Relative accuracy measures for stroke gestures.
Proceedings of the ACM International Conference on Multimodal Interfaces (ICMI '13).
Sydney, Australia (December 9-13, 2013).
New York: ACM Press, pp. 279-286.
Vatavu, R.-D., Anthony, L. and Wobbrock, J.O. (2012).
Gestures as point clouds: A $P recognizer for user interface prototypes.
Proceedings of the ACM International Conference on Multimodal Interfaces (ICMI '12).
Santa Monica, California (October 22-26, 2012).
New York: ACM Press, pp. 273-280.
Anthony, L. and Wobbrock, J.O. (2012).
$N-Protractor: A fast and accurate multistroke recognizer.
Proceedings of Graphics Interface (GI '12).
Toronto, Ontario (May 28-30, 2012).
Toronto, Ontario: Canadian Information Processing Society, pp. 117-120.