Multimodal human detection and fuzzy decisional engine for interactive behaviors of a mobile robot

The human detection is an essential function for personal robots that are supposed to be useful as cognitive communication and information systems. In this paper we describe a method for reliable detection that presents two contributions: 1. the multimodal approach to improve the punctual detection; 2. the synthesis of the robot behaviors for reliable tracking by keeping the person in the robot field of view. To improve the punctual detection rate we propose the combination of the outputs of different low-cost sensors embedded in a robot: laser, RGB cameras, IR camera and microphones. The fusion of the results of four detectors is made taking into account spatial criteria according to the field of view and to the performances of each sensor. We demonstrate experimentally that the designed reasoning system is able to improve the human detection in complex situations that may lead the system to lose the person detection because of important occlusions. The usefulness of the framework is shown also in the case of face-to-face communication: the robot focuses the human in order to improve the quality of the communication.