Estimation of the number of states for gesture recognition with Hidden Markov Models based on the number of critical points in time sequence

This paper presents a method of choosing the number of states of a Hidden Markov Model (HMM) based on the number of critical points in motion capture data. The choice of HMM parameters is crucial for recognizer’s performance as it is the first step of the training and cannot be corrected automatically within an HMM. In this article we define the predictor of the number of states based on the number of critical points of a sequence and test its effectiveness against sample data from the IITiS Gesture Database.

► We propose a method of choosing number of HMM states for motion capture data detection.
► Proposed predictor is based upon number of critical points of detected sequence.
► We test predictor on motion capture data against Akaike Information Criterion.
► Choosing number of states equal to predictor improves value of AIC for finger sensors.
► The effect is less significant for accelerometers data.