Generating datasets of human activities in pervasive spaces when using wearable devices

Abstract

Reliability and accuracy have always been the centre point of data collection. It is not always possible to collect all necessary data required for research. Various researchers identify this aspect as a challenge in their work. Most cases occur when data is related to human activates, such when recording location or patterns for a specific amount of time for a large sample group. This work presents PacSim (Pervasive Activity Context Simulator), a first prototype to investigate the possibility of generating data based on pervasive reconstruction of observed activities. The aim for this simulator is to generate datasets of human activities based on datasets gathered in a real life environment in order to help the researcher by filling in any missing or required data. In addition, a prototype is proposed for a wearable device which will collect the activity datasets required by the simulator. This device will recognize human activities data gathered by a single accelerometer and an envelope matching algorithm to identify previously fingerprinted activities. The device is worn on the subject’s dominant hand and will form part of a context aware system, where location is also related to the recognition of the activity. Through an experiment spread over three days at Saint Vincent de Paul, four patients were recruited, where the datasets required were collected by first, logging data manually by annotating activities and secondly, by logging data through the wearable device designed purposely for this project. PacSim replicated the first day using an algorithm described in this thesis, where the activities generated were evaluated in comparison to the actual activities performed by the subject.