Persuasive Agents

The goal of the “Persuasive Agents”-project is to develop novel techniques for persuading the members of a household to save energy. To this purpose teams of psychologists (Eindhoven) and computer scientists (Tilburg) collaborate with Smart Homes, the Dutch Expertise Centre on Home Automation & Smart Living.

Visits today:4
Visits in total:21247
8 years ago
7 years ago
7 years ago

Contact Info

R. Mattheij MA
Room D 336
PO Box 90153
5000 LE Tilburg
31 13 466 2787
Tilburg University


I am a PhD student at Tilburg University. My field of expertise is the development of personalized agent systems that aim to persuade people to change their energy-consumption behavior.

I am now working on "the eyes" of such agent systems, by exploring possibilities to improve automatic face detection methods. Automatic face detection from videos is an active research area in the domain of artificial intelligence. State-of-the-art face-detection algorithms rely on Viola-Jones classifiers, dedicated machine-learning algorithms that are trained on visual images of faces and non-faces. Although these algorithms are capable of detecting faces in real time, their detection performance is far from perfect. We aim to improve face-detection performance by training Viola-Jones classifiers on depth images acquired by the Microsoft Kinect. We expect depth information to complement visual information, because, unlike visual images, depth images are more or less invariant to changes in illumination conditions. The research question addressed in our study is: to what extent do depth images improve the face detection performance of Viola-Jones classifiers? To answer this question, we are compiling an annotated dataset of depth and visual images of faces and non-faces under various illumination conditions and for different views of the faces. In our experiments, we will evaluate the performances of Viola-Jones classifiers trained on three subsets of data: (1) visual only, (2) depth only, and (3) visual-dept combined. The results will reveal how visual and depth data contribute to the face-detection performance.

Our face-detection study is the first step in the development of software that generates a virtual person displayed on a screen. The virtual person can engage in a “nonverbal dialogue” with humans standing in front of the screen. Once we have endowed our virtual person with improved face and eye detection, it can eye-gaze at humans. Subsequently, our future research will focus on person identification and expression recognition to allow the virtual person to respond appropriately to humans and to human nonverbal expressions.

Our virtual person will be used in behavioral experiments (performed by our project partners) to determine if virtual persons can persuade members of a household to lower their energy consumption.


Mattheij, R.J.H. & Postma, E.O. (2012). The eyes have it: towards enhancing sustainability. Proceedings of the 6th International Conference on Persuasive Technology (Persuasive 2011), Columbus, OH, USA, 2012. (In press)

Mattheij, R.J.H., Szilvasi, L., Beer, L. de, Rakiman, K. & Shahid, S. (2011). GooGreen: Towards Increasing the Environmental Awareness of Households. In J.A. Jacko (Ed.), Human-Computer Interaction, Part III, HCII 2011 (pp. 500-509). Hilton Orlando Bonnet Creek, Heidelberg: Springer

Role in project

PhD student