Multimodal Identification of Journeys

Due to the ubiquity of localisation technology, users now have the ability to keep a record of their own location, as a kind of `location diary'. Such a large collection of data can become unmanageable without some way to structure that data to make it useful and searchable. We address this problem of structuring location data by proposing a framework for classifying the data into often-traversed routes. In this work, commonly traversed routes are identified with clusters based on sensed data. Our framework does not rely on any one source of location information, but can fuse data from multimodal localisation sources. We demonstrate the effectiveness of our algorithm by examining the combination of GPS, wireless signal strength readings and image-matching on very challenging data in a variety of environmental conditions. By fusing these three modalities we obtained better performance than any individual or combination of two modalities. As it can be orientated towards the needs and capabilities of the user based on context, this method becomes useful for some ambient assisted living applications.