Role-based and individual personalization in museums with pre-trained computer vision models

This paper provides first insights into using pretrained deep learning models for computer vision to support lightweight personalization in museums without collecting data about visitors. It presents a prototype for embodied interaction with projection-mapped Regency interiors, which offers role-base personalization based on avatars and human pose estimation, and individual personalization based on personal portraits and artistic style transfer. An empirical evaluation with museum visitors found that these personalization approaches helped visitors to relate to the Regency interior they engaged with and contributed to a positive user experience. Privacy concerns around processing live camera images with deep learning models were mitigated by participants' high levels of trust in the museum to use their data ethically. The findings are relevant for researchers and practitioners exploring new ways of personalization in museums.