Mobile motion sensor-based human activity recognition and energy expenditure estimation in building environments

Abstract

This paper presents a work on human activity recognition (HAR) using motion sensors embedded in a smart phone in building environments. Our HAR system recognizes general human activities including walking, going-upstairs, going-downstairs, running, and motionless, using statistical and orientation features from signals of motion sensors and a hierarchical Support Vector Machine classifier. Upon activity recognition, our system also generates energy expenditures of the recognized physical activities: energy expenditures are computed based on Metabolic Equivalents (METS) values, step count, distance, speed, and duration of activities. By testing our system in building environments, we have obtained an average recognition rate of 98.26% with physically consumed energy information. With the presented system, different building designs and environments can be evaluated in terms of energy consumptions of residents for their physical activities. © Springer-Verlag Berlin Heidelberg 2013.