Home // International Journal On Advances in Software, volume 8, numbers 1 and 2, 2015 // View article

Car Drive Classification and Context Recognition for Personalized Entertainment Preference Learning

Authors:
Thomas Christian Stone
Stefan Haas
Sarah Breitenstein
Kevin Wiesner
Bernhard Sick

Keywords: Context-aware Vehicle; Spatial Clustering; Drive Context Prediction; In-Car Infotainment; Automation

Abstract:
The automotive domain, with its increasing number of comfort and infotainment functions, offers a field of opportunities for pervasive and context-aware personalization. This can range from simple recommendations up to fully automated systems, depending on the information available. In this respect, frequent trips of individual drivers provide promising and interesting features, on the basis of which, usage patterns may possibly be learned and automated. This automation of functions could increase safety as well as comfort, as the driver can concentrate more on the experience of driving instead of repeatedly and manually adjusting comfort- or entertainment-related systems. To identify frequent driving contexts in a set of recorded signal in a vehicle, e.g., GPS tracks, this paper presents two different clustering algorithms: First, a hierarchical Drive-Clustering, which combines drives based on their number of common GPS points. Second, a Start-Stop-Clustering, which combines trips with the same start- and stop-cluster utilizing density based clustering. The Start-Stop-Clustering showed particularly good results, as it does not depend on the concrete routes taken to a stop position and it is still able to detect more trip clusters. To predict these drives, a Bayesian network is presented and evaluated, with logged trip data of 21 drivers. The Bayes Net uses context information, i.e., the time, weekday and the number of people in the car, to predict the most likely drive context with high accuracy. A new automated entertainment source selection algorithm demonstrates the usefulness of the retrieved information. The algorithm learns and predicts a driver's preferences for selected entertainment sources depending on recognized drive contexts.

Pages: 53 to 64

Copyright: Copyright (c) to authors, 2015. Used with permission.

Publication date: June 30, 2015

Published in: journal

ISSN: 1942-2628