Depth camera driven mobile robot for human localization and following,
Abstract
In this paper the design and the development of a mobile robot able to locate and then follow a human target is described. Both the integration of the required mechatronics components and the development of appropriate software are covered. The main sensor of the developed mobile robot is an RGB-D camera. This sensor is usually combined with the appropriate middleware that can detect humans in a scene and also provide the position of the detected human in the 3D space. One of the cues this middleware’s algorithms are using to detect humans is motion, thus resulting in many false detections when applied to data captured by a mobile platform. This work proposes the use of a special-tailored feed forward neural network to further process the initial detections, identifying and rejecting most false positives. Experimental results based on two self-captured data sets show the improved detection rate of our two-stage human detector, as well as the efficient real-time performance of the proposed system for human localization and following.