A Low-power Real-time Hidden Markov Model Accelerator for Gesture User Interface on Wearable Devices

A low-power and real-time hidden Markov model (HMM) accelerator is proposed for gesture user interface on wearable smart devices. HMM algorithm is widely used for sequence recognition problems such as speech recognition and gesture recognition thanks to its best-in-class recognition accuracy. However, the HMM algorithm has high computational complexity and requires massive memory bandwidth in sequence matching process. Therefore, there have been studies on hardware acceleration of the HMM algorithm to resolve these issues, but they were focusing on the speech recognition and therefore did not accommodate the motion orientation function required for the gesture recognition problem. The motion orientation function computes the direction of hand movement in gesture sequence and thus involves compute intensive division and arctangent operations. In this paper, we propose an HMM accelerator with a light weight motion orientation module for realizing gesture recognition on wearable devices. Binary search method is exploited in the motion orientation module to avoid the division and arctangent operations associated with calculating orientations for reduced arithmetic complexity. In addition, gesture models are clustered in the gesture database to reduce external memory transactions. Moreover, logarithmic arithmetic is adopted in Viterbi decoder of the HMM algorithm for more reduction in its complexity. Thanks to these proposed schemes, this work achieves 25.6% power reduction compared with a vanilla hardware implementation of the gesture recognizing HMM.