Segmentation of the Human Gait Cycle Using Hidden Markov Models (HMM)
Abstract
This paper provides a supervised Hidden Markov Model (HMM) for the segmentation of the human gait cycle. The model arises, in turn, from the combination of two HMM models, each with three hidden states, making use of the DepmixS4 and RcppHMM libraries of the free software R. The validation of the model was carried out with the cross-validation method in two different ways. The three accelerometer signals provided by the sensor located in the left ankle and in the right ankle, respectively, were processed in the 20 healthy young subjects (33.4 +/- 7 years, height 172.6 +/- 9.5 cm, muscle mass 73.2 +/- 10.9 kg), the open base MAREA. The base has different tests in indoor and outdoor environment, allowing a variety of walking situations, even the combination of walking and running. In this way the model provides a new validation for the base. The results were expressed from the statistics derived from the confusion matrix: Accuracy, Sensitivity and Specificity. In the tests of walking for 3 min on the flat surface in close environment, the model reached: 99%, 84.1% and 93.2% respectively (sensor on left ankle). With the signals obtained from the right ankle, the valueswere 93%, 73% and 86.4%, respectively. In both cases, the acceleration signals were filtered with the Butterworth filter. The results are discussed with other authors who have used the same base with different algorithms.
