A Hierarchical Approach towards Activity Recognition

Activity recognition with the use of inertial sensors, namely accelerometers and gyroscopes, has gained increasing attention during the last decades. In this work, we propose a novel way of tackling activity classiffication by developing a multi-step hierarchical classiffication algorithm. While previous research has looked at the problem as a whole, by adopting one of the two major approaches for activity recognition - the sliding window approach and primitive-based approach, our system will divide the classiffication problem into smaller classiffication problems following a hierarchical approach for improve on accuracy and computational cost. This work aims at detecting self-neglect behaviour in a living environment. As such, the activities chosen to be classiffied consist of quotidian daily living activities such as walking, brushing teeth, washing hands, typing at the computer, sitting, stand and picking up something from the floor. The experimental work has shown promising results which support the use of the multi-step hierarchical approach proposed in this paper.