Detecting changes in motion characteristics during sports training.

This paper proposes a stochastic approach for representing and analyzing the gradual changes that occur in human movement during sports training. Human movement primitives are described using Factorial Hidden Markov Models, and compared using the Kullback-Liebler distance, a measure of information divergence between two models. This representation is combined with an automated segmentation and clustering approach to enable the system to autonomously extract and group together movement primitives from continuous observation of human movement data. The proposed system is tested on a human movement dataset obtained over 4 months during training for a marathon. Experimental results demonstrate that the system is able to detect gradual changes in the human movement.