Centinela: A human activity recognition system based on acceleration and vital sign data

摘要: This paper presents Centinela, a system that combines acceleration data with vital signs to achieve highly accurate activity recognition. Centinela recognizes five activities: walking, running, sitting, ascending, and descending. The includes portable unobtrusive real-time collection platform, which only requires single sensing device mobile phone. To extract features, both statistical structural detectors are applied, two new features proposed discriminate among activities during periods of sign stabilization. After evaluating eight different classifiers three time window sizes, our results show achieves up 95.7% overall accuracy, is higher than current approaches under similar conditions. Our also indicate useful between certain activities. Indeed, 100% accuracy for such as running slightly improves the classification ascending compared cases utilize only.