Environment feature extraction and classification for Context aware Physical Activity monitoring

Abstract

Context aware Physical Activity (PA) monitoring of humans is important for the study of diseases associated with obesity and lack of physical activity. This paper introduces a wearable context aware PA monitoring device which determines if the user is indoors or outside in situations of disrupted Global Positioning System (GPS) reception. In addition to a GPS sensor, multiple light and temperature sensors were added to our PA monitoring device. Differences in inside and outside temperature and the intensity of light are used to distinguish the context of location. Location, Light and temperature values were recorded using a controlled route during a period of 20 days in January and February. One of the non-parametric pattern recognition techniques (K-nearest neighbors) was used to classify indoor and outdoor conditions based on the combination of sensor values. Results show that the K-nearest neighbors algorithm could distinguish indoor and outdoor conditions during daytime and nighttime with the error of 0.003.