Gesture recognition using Markov Systems and wearable wireless inertial sensors

Abstract

Wearable wireless devices and ubiquitous computing are expected to grow significantly in the coming years. Standard inputs such as a mouse and keyboard are not well suited for such mobile systems and gestures are seen as an effective alternative to these classic input styles. This paper examines gesture recognition algorithms that use an inertial sensor worn on the forearm. The recognition algorithms use the sensor's quaternion orientation in either a Hidden Markov Model or Markov Chain based approach. A set of six gestures were selected to fit within the context of an active video game. Despite the fact that the Hidden Markov Model is one of the most commonly used methods for gesture recognition, the experiments showed that the Markov Chain based algorithms outperformed the Hidden Markov Model. The Markov Chain algorithm obtained an average accuracy of 95%, while also having a much faster computation time, making it better suited for real time applications.