Tangible User Interfaces (TUIs) feature physical objects that people can manipulate to interact with smart spaces. Smart objects used as TUIs can further improve user experience by recognizing and coupling natural gestures to commands issued to the computing system. Hidden Markov Models (HMM) are a typical approach to recognize gestures sampled from inertial sensors. In this paper we implement a HMM-based continuous gesture recognition algorithm, optimized for low-power, low-cost microcontrollers without floating point unit. The proposed solution is validated on a set of gestures performed with the Smart Micrel Cube (SMCube), which embeds a 3-axis accelerometer and an 8-bit microcontroller. Through the paper we evaluate the implementation issues and describe the solutions adopted for gesture segmentation and for the fixed point HMM forward algorithm. Furthermore, we explore a multiuser scenario where up to 4 people share the same device. Results show that the proposed solution performs comparably to the standard forward algorithm and can be efficiently used for low cost smart objects.
Continuous Gesture Recognition for Resource Constrained Smart Objects
Milosevic, Bojan;Farella, Elisabetta;
2010-01-01
Abstract
Tangible User Interfaces (TUIs) feature physical objects that people can manipulate to interact with smart spaces. Smart objects used as TUIs can further improve user experience by recognizing and coupling natural gestures to commands issued to the computing system. Hidden Markov Models (HMM) are a typical approach to recognize gestures sampled from inertial sensors. In this paper we implement a HMM-based continuous gesture recognition algorithm, optimized for low-power, low-cost microcontrollers without floating point unit. The proposed solution is validated on a set of gestures performed with the Smart Micrel Cube (SMCube), which embeds a 3-axis accelerometer and an 8-bit microcontroller. Through the paper we evaluate the implementation issues and describe the solutions adopted for gesture segmentation and for the fixed point HMM forward algorithm. Furthermore, we explore a multiuser scenario where up to 4 people share the same device. Results show that the proposed solution performs comparably to the standard forward algorithm and can be efficiently used for low cost smart objects.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
