People Detection by Mobile Robots Doing Automatic Guard Patrols

Abstract

Patrolling and surveillance of buildings and other human-made environments is among dull and sometimes dangerous tasks that can be automated by teams of security robots. A key feature of such robots is the ability to detect and track people, as the event of detecting a person is often related with unauthorized presence of people in specific places. This paper presents a perception system to endow mobile robots with the ability of detecting people while patrolling continuously a facility. The system merges the output of four state-of-the-art detectors having different capabilities regarding distance of detection and sensor modality used, so as to exploit their complementary features and provide more reliable people detection. After assessing experimentally the strengths and limitations of the individual detectors, a fuzzy logic-based frame rate adaptive controller is presented to assign, at each iteration, computation resources to each detector depending on the perception situation in front of the robot. The system was demonstrated successfully and validated in experiments carried out in a hallway-like area of a real office facility.