Abstract

Tracking and recognizing people are essential skills modern service robots have to be provided with. The two tasks are generally performed independently, using ad-hoc solutions that first estimate the location of humans and then proceed with their identification. The solution presented in this paper, instead, is a general framework for tracking and recognizing people simultaneously with a mobile robot, where the estimates of the human location and identity are fused using probabilistic techniques. Our approach takes inspiration from recent implementations of joint tracking and classification, where the considered targets are mainly vehicles and aircrafts in military and civilian applications. We illustrate how people can be robustly tracked and recognized with a service robot using an improved histogram-based detection and multisensor data fusion. Some experiments in real challenging scenarios show the good performance of our solution.