Data Reduction Methods for Life-Logged Datasets

Abstract

Life-logging utilises sensor technology to automate the capture of a person’s interaction with their environment. This produces useful information to assess wellbeing, but this information is often buried within the volume of data. In this chapter, we analyse a life-log comprising image data and contextual information and apply algorithms to collate, mine and categorise the data. Four approaches were investigated: (i) Self-reporting of important events by the person who collected the data; (ii) Clustering of images into location-based events using GPS metadata, (iii) Face detection within the images and (iv) Physiological monitoring using Galvanic Skin Response (GSR); as a way to identify more meaningfulimages. Using a bespoke wearable system, comprising a smartphone and smartwatch, six healthy participants recorded a life-log in the form of images of their surroundings coupled with metadata in the form of timestamps, GPS locations, accelerometer data and known social interactions. Followingapproximately 2.5 h of recording, the data reduction methodologies outlined above were applied to each participant’s dataset, yielding an 80–86% reduction in size which facilitates more realistic self quantification. However, each approach has some shortcoming and the data reduction method used will need personalisation and depend on the intended application.