Accelerometer-based Gesture Recognition Using MFCC and HMM

Abstract

In this paper, an accelerometer-based gesture recognition system for mobile devices interaction has been proposed. A new feature extraction method for gesture recognition based on Mel Frequency Cepstrum Coefficient (MFCC) has been put forward. Since the accelerometer signals are sensitive to device's rotation, firstly a resultant force of acceleration is obtained and then their MFCCs are extracted as features. The Classifier used is Hidden Markov Model(HMM). An important problem is how to choose initial estimates of the HMM parameters. In our work, the segmental k-means segmentation with clustering method is adopted to estimate initial model parameters. The average recognition result of twenty complex gestures using the proposed method is effective. The experimental results show that gesture-based interaction can be used as a novel human computer interaction for consumer electronics and mobile devices.