Human activity recognition based on novel accelerometry features and hidden markov models application

The Human Activity Recognition (HAR) systems require objective and reliable methods that can be used in the daily routine and must offer consistent results according to the performed activities. In this work, a framework for human activity recognition in accelerometry (ACC) based on our previous work and with new features and techniques was developed. The new features set covered wavelets, the CUIDADO features implementation and the Log Scale Power Bandwidth creation. The Hidden Markov Models were also applied to the clustering output. The Forward Feature Selection chose the most suitable set from a 423th dimensional feature vector in order to improve the clustering performances and limit the computational demands. K-means, Affinity Propagation, DBSCAN and Ward were applied to ACC databases and showed promising results in activity recognition: from 73:20%±7:98% to 89:05%±7:43% and from 70:75%±10:09% to 83:89%±13:65% with the Hungarian accuracy (HA) for the FCHA and PAMAP databases, respectively. The Adjust Rand Index (ARI) was also applied as clustering evaluation method. The developed algorithm constitutes a contribution for the development of reliable evaluation methods of movement disorders for diagnosis and treatment applications.