A system including at least three robots. Each robot including a proximity sensor unit and an imaging device. At least one robot including a processor to perform a method of estimating a pose of a human, the method including obtaining a first pose estimate for the human, the first pose estimate based on proximity sensor information, obtaining a second pose estimate for the human, the second pose estimate based on imaging device information, and generating a refined pose estimate for the human by fusing the first pose estimate with the second pose estimate, where the first pose information provides predictive values and the second pose estimate provides correction values. The method including applying a deep neural network (DNN) human model, and applying a DNN human pose model. A method to generate a refined pose estimation for a human and a non-transitory computer readable medium are also disclosed.