Monitoring and estimating physical activity (PA) can be useful in the prevention and treatment of obesity or aging-related disorders, for example. Even if there exist some reliable methods to evaluate the level of physical activity (such as oxygen uptake measurement or doubly labeled water), those are often expensive and intrusive and then do not suit for daily use. An alternative approach for the assessment of PA involves the use of unconstrained wearable systems such as accelerometers. The problem of PA and energy expenditure (EE) estimation from accelerometer signals has received much attention for the latter years. Since the posture and the nature of the movements involved in different types of PA strongly affect the EE, quantitative information (raw accelerometer data) is not sufficient to efficiently assess EE and some additional and qualitative labelling is often needed. The final aim of this study is to identify a subject's PA behaviour (namely, the postures and activities performed throughout the day) from the accelerometer data in order to precisely estimate the EE related to the PA. The process can therefore be summarized as a segmentation/classification task, where each sample or frame is to be labelled with one posture or activity label. The algorithm presented here identifies three dynamic activities (biking, walking and running) from data recorded on the shin and a second system dedicated to walking periods detects changes in speed and in incline. The methods are tested on a 24-subject corpus with data acquired in controlled conditions.