Non-intrusive and Privacy Preserving Activity Recognition System for Infants Exploiting Smart Toys

The Human Activity Recognition (HAR) research area showed great advances in the last decade, achieving excellent prediction performances and great applicability, which is reflected on the wearable sensors market adoption. However, most of the research effort concentrated on an adult target population. When considering a younger population of infants or children, currently available HAR solution based on wearable devices are not applicable anymore. In this paper we present an HAR based solution targeting infants, based on a non-intrusive and privacy-preserving measurement methodology which allows the preservation of children behaviour and the collection of objective data (particularly important for clinical observation purposes). The proposed solution, based on the usage of a set of smart toys (AutoPlay toys-set) achieves great performances in the recognition of a set of 12 toy-activity pairs, reaching accuracy values up to 96%. These results pave the way to a broad application of the presented methodology on objective analysis of humans motor skills.