Continuous Gesture Recognition Based on Hidden Markov Model

TLDR: A gesture recognition based on accelerometer, which is modeled by Hidden Markov Model, is proposed, which works well in detecting valid gesture data while recognition time and the computation load can be reduced in the case of guaranteeing recognition precision.

Abstract: Gesture is a compelling interactive mode, which makes interaction become more active than before. With the development of acceleration sensor, it has played an important role in gesture recognition of human-computer interaction. This paper represents a gesture recognition based on accelerometer, which is modeled by Hidden Markov Model (HMM). For “continuous” gesture recognition, it is a vital problem of how to obtain real valid data in a series of raw gesture data accurately and efficiently. To solve this, we proposed a new gesture detection method based on energy entropy and combined with threshold. Gesture data is analyzed in energy distribution of frequency domain by Short Time Fourier Transform (STFT), which can calculate energy entropy that reflects signal energy distribution. Then an appropriate threshold is set up to determine the start and end of gesture. Through experiments, the proposed method can be proved that it works well in detecting valid gesture data while recognition time and the computation load can be reduced in the case of guaranteeing recognition precision.