Positioning and activity recognition for context awareness

Abstract

This study tries to utilize positioning (location detection) and activity recognition to achieve context awareness by only using tools and sensors that are standard in modern smart phones. User locations are gathered through participatory crowdsourcing, where users help in identifying locations and landmarks. Data gathered from crowdsourcing is then coupled with user activity data that are gathered by using accelerometers that are already embedded in the user’s mobile devices. We attempted to create a Hidden Markov Model (HMM) that is capable in identifying current user context and accurately predict user’s immediate future contexts using inputs that consist of users’ locations and users’ activities. Given certain location, these inputs were in the form of continuous, sequential activities. Accuracy rates of the context awareness using the above parameters as well as the HMM model are shown to be promising with percentage rates reaching 90-95% for some contexts. Such accuracy rates are possible because HMM models are inference models that are well suited for input data that are sequential in nature. The study concludes, as initially hypothesized, that it is possible to utilize positioning and activity recognition to create a HMM model that accurately show users’ context as well as predicting future context.