A graph based algorithm for postures estimation based on accelerometers data

An algorithm is presented in this paper to estimate the activity of a person at each time index through a Bayesian approach. It does not require any pre-processing to identify single activity parts in the signal but can generate some instability in the estimated sequences of activities due to its nature. To circumvent this issue, the decision is enforced with a graph setting constraints between the different activities to be detected. As the method is based on a Bayesian approach, training can be used to adapt the algorithm to each person. If so, the algorithm shows very good performance.