Paper Reading 19 – A $3 Gesture Recognizer – Simple Gesture Recognition for Devices Equipped with 3D Acceleration Sensors

Comments
Patrick Frith
Paola Garza

Reference Information
A $3 Gesture Recognizer – Simple Gesture Recognition for Devices Equipped with 3D Acceleration Sensors
 - Sven Kratz and Michael Rohs
 - IUI 2010, Hong Kong, China

Summary
The $3 Gesture Recognizer is targeted at providing an easy to use and easy to implement gesture recognizer to be used in prototyping environments.  Essentially, for designing gesture-based applications, this device could be used as an easy effective alternative to more expensive and complex solutions.  This device is also based on simple trigonometric and geometric calculations, so no external frameworks or toolkits are needed in order to use it.

A main issue with other gesture recognition devices is that they are mostly based on a 2-D landscape.  If a gesture cannot be mapped to a 2-D plain, then it cannot be included in the device’s gesture subset.  The designers set out to expand on the work of Wobbrock et al by implementing a 3-D version of their device.  They also point out how their work is similar to how Nintendo’s WiiMote functions and recognizes gestures, and they end up using the WiiMote to capture gestures.

Their user study involved 12 participants performing each gesture in the diagram of the subset 15 times.  On average, their implementation of gesture recognition resulted in an 80% recognition rate.  In the figure included, (b) had the highest recognition rate, while (j) caused the most trouble for users.  They explain this might be due to the ambiguity of the motion (j) requires and how wide open to interpretation that is to the user.  Users also found (h) and (i) the most uncomfortable to perform.

Overall, they only had an 8% false positive rate.  Although, they did point out some weaknesses.  Each gesture must be explicitly started and stopped by the user.  The recognizer is unable to pick up on gestures in the middle of other motions.  They also are only able to store so many gestures in the subset.  If the subset becomes too large, false positives rise and there is too much overhead to calculate the recognition.  However, the test was a success, and they were able to produce an alternative to more advanced recognition approaches that is cheaper and more convenient to use.

Discussion
Gesture-based applications have always been fun to use.  I’ve only really used the Wii myself, but the gesture recognition in devices like that is intriguing.  I never put much though into the science behind this recognition until reading this paper though.  It seems like they have a pretty useful idea they are developing.  While most of my ideas point to uses in gaming, this technology could be used for many different things.  One of these could surgeon training or some other training that requires great motor skills and coordination.

But, they have already stated how they have a limited subset of gestures.  I imagine something like a surgeon trainer would require extremely accurate gesture recognition techniques.  I am sure the $3 Gesture Recognizer could be up to the task.  They would just need to improve its technique so it could calculate far more gestures than it currently can.  I look forward to seeing where gesture-based technologies go in the near and far future.