SPENCER: A Socially Aware Service Robot for Passenger Guidance and Help in Busy Airports
In this paper we introduce video features which are used to predict if people want to exchange contact information with the other in a speed-date, we also use these features to predict how physically attractive participants found their dates. Previous work on predicting and interpreting speed-dates has focused mainly on the audio channel. We use automatically extracted features related to position, proximity and motion. This paper shows that these features can be used to significantly outperform the baseline and have comparable performance to audio-only systems. The data used has been gathered from a real speed-date event, involving 16 participants. Experiments were carried out on 64 speed-dates lasting 5 minutes. The best performance on prediction exchanging contact information was 72% and 70% accuracy for males and females respectively, and 70% for both genders when predicting physical attraction.