Human pose fusion for scalable kinetic interfaces

In this paper we present a method for fusion of human poses that are estimated simultaneously from multiple RGB-D cameras. The proposed solution is scalable in terms of the number of used sensors, processing power, network overhead and keeping latency within small temporal bounds. Spatial relations of cameras are continuously estimated in near real-time from the acquired data. Sensors can be added, removed or repositioned without significant disruption of the systems functioning.