Activity classification at a higher level: what to do after the classifier does its best?

Abstract

Research in activity classification has focused on the sensors, the classification techniques and the machine learning algorithms used in the classifier. In this work, we study a higher level of activity classification. We present two methods that can take the final observations of a classifier and improve them. The first method uses hidden Markov models to define a probabilistic model that can be used to improve classification accuracy. The second method is a novel method that we developed that uses probabilistic models along with matching costs in order to improve accuracy. Testing showed that both proposed methods presented a significant increase in classification accuracy rates, while also proving that they can both run in real time.