On-line classification of human activity and estimation of walk-run speed from acceleration data using support vector machines

The awareness of the physical activity that human subjects perform, and the quantification of activity strength and duration are important tasks that a wearable sensor system would fulfill to be valuable in several biomedical applications, from health monitoring to physical medicine and rehabilitation. In this work we develop a wearable sensor system that collect data from a single thigh-mounted tri-axial accelerometer; the system performs activity classification (sit, stand, cycle, walk, run), and speed estimation for walk (run) labeled data features. These classification/estimation tasks are achieved by cascading two Support Vector Machines (SVM) classifiers. Activity classification accuracy higher than 99% and root mean square errors E_RMS = 0.28 km/h for speed estimation are obtained in our preliminary experiments. The developed wearable sensor system provides activity labels and speed point estimates at the pace of two readings per second.