Event-driven feature analysis in a 4D spatiotemporal representation for ambient assisted living

This paper presents a detailed analysis of a 4D representation of events, which are generated by a dynamic stereo vision sensor for the recognition of person´s fall. Dynamic vision detectors consist of self-signaling pixels that autonomously react to scene dynamics and asynchronously generate events upon relative light intensity change. Their complete on-chip redundancy reduction, wide dynamic range and high temporal resolution allow efficient and continuous activity monitoring in natural environment. Using a stereo pair of dynamic vision detectors, it is possible to represent the scene dynamics in a 4D space (including time) at a high temporal resolution. In this work, we performed 100 recordings of scenarios including falls in indoor environment using this dynamic stereo vision sensor. Seven features have been extracted and analyzed for three types of falls such that robust parameters will be kept for fall recognition. The result of this analysis is shown in this work with promising outcomes.