On the exploitation of Hidden Markov models to improve location-based temporal segmentation of egocentric videos