Finding Motifs in Large Personal Lifelogs

The term Visual Lifelogging is used to describe the process of tracking personal activities by using wearable cameras. A typical example of wearable cameras is Microsoft's SenseCam that can capture vast personal archives per day. A significant challenge is to organise and analyse such large volumes of lifelogging data. State-of-the-art techniques use supervised machine learning techniques to search and retrieve useful information, which requires prior knowledge about the data. We argue that these so-called rule-based and concept-based techniques may not offer the best solution for analysing large and unstructured collections of visual lifelogs. Treating lifelogs as time series data, we study in this paper how motifs techniques can be used to identify repeating events. We apply the Minimum Description Length (MDL) method to extract multi-dimensional motifs in time series data. Our initial results suggest that motifs analysis provides a useful probe for identification and interpretation of visual lifelog features, such as frequent activities and events.