A light computing method for real-time activity recognition

Accelerometers offer a non-intrusive and low-cost solution to quantify physical activity (PA) level, but most commercialised systems provide a rough approximation based on activity counts. Moreover, few of them enable PA recognition during daily life, which is an important issue to evaluate quality of life. Based on a simple PA classification, some previous works (Bussmann et al. 2001) assessed postures and motions during daily life by means of accelerometers. Unfortunately, they do not yet allow to evaluate a large population in real time. In this work, we propose a new method that minimises the data flow transmission in order to set up an ambulatory monitoring device that could be integrated into a sensor network. This study is focused on postures and locomotion activities recognition and on velocity prediction.