Nicholas Gillian, Sara Pfenninger, Spencer Russell, Joseph A. Paradiso

Abstract

Gestures Everywhere is a dynamic framework for multimodal sensor fusion, pervasive analytics and gesture recognition. Our framework aggregates the real-time data from approximately 100 sensors that include RFID readers, depth cameras and RGB cameras distributed across 30 interactive displays that are located in key public areas of the MIT Media Lab. Gestures Everywhere fuses the multimodal sensor data using radial basis function particle filters and performs real-time analysis on the aggregated data. This includes key spatio-temporal properties such as presence, location and identity; in addition to higher-level analysis including social clustering and gesture recognition. We describe the algorithms and architecture of our system and discuss the lessons learned from the systems deployment.