Recognizing human activities from accelerometer and physiological sensors

Recently the interest about the services in the ubiquitous environment has increased. These kinds of services are focusing on the context of the user's activities, location or environment. There were many studies about recognizing these contexts using various sensory resources. To recognize human activity, many of them used an accelerometer, which shows good accuracy to recognize the user's activities of movements, but they did not recognize stable activities which can be classified by the user's emotion and inferred by physiological sensors. In this paper, we exploit multiple sensor signals to recognize user's activity. As Armband includes an accelerometer and physiological sensors, we used them with a fuzzy Bayesian network for the continuous sensor data. The fuzzy membership function uses three stages differed by the distribution of each sensor data. Experiments in the activity recognition accuracy have conducted by the combination of the usages of accelerometers and physiological signals. For the result, the total accuracy appears to be 74.4% for the activities including dynamic activities and stable activities, using the physiological signals and one 2-axis accelerometer. When we use only the physiological signals the accuracy is 60.9%, and when we use the 2 axis accelerometer the accuracy is 44.2%. We show that using physiological signals with accelerometer is more efficient in recognizing activities.