An inertial measurement unit for user interfaces

Abstract

Inertial measurement components, which sense either acceleration or angular rate, are being embedded into common user interface devices more frequently as their cost continues to drop dramatically. These devices hold a number of advantages over other sensing technologies: they measure relevant parameters for human interfaces and can easily be embedded into wireless, mobile platforms. The work in this dissertation demonstrates that inertial measurement can be used to acquire rich data about human gestures, that we can derive efficient algorithms for using this data in gesture recognition, and that the concept of a parameterized atomic gesture recognition has merit. Further we show that a framework combining these three levels of description can be easily used by designers to create robust applications. A wireless six degree-of-freedom inertial measurement unit (IMU), with a cubical form factor (1.25 inches on a side) was constructed to collect the data, providing updates at 15 ms intervals. This data is analyzed for periods of activity using a windowed variance algorithm, whose thresholds can be set analytically. These segments are then examined by the gesture recognition algorithms, which are applied on an axis-by-axis basis to the data. The recognized gestures are considered atomic (i.e. cannot be decomposed) and are parameterized in terms of magnitude and duration. Given these atomic gestures, a simple scripting language is developed to allow designers to combine them into full gestures of interest. It allows matching of recognized atomic gestures to prototypes based on their type, parameters and time of occurrence. Because our goal is to eventually create stand-alone devices,the algorithms designed for this framework have both low algorithmic complexity and low latency, at the price of a small loss in generality. To demonstrate this system, the gesture recognition portion of (void*): A Cast of Characters, an installation which used a pair of hand-held IMUs to capture gestural inputs, was implemented using this framework. This version ran much faster than the original version (based on Hidden Markov Models), used less processing power, and performed at least as well.