Robust and accurate global vision system for real time tracking of multiple mobile robots

This paper presents a new global vision system for tracking of multiple mobile robots. To the best knowledge of the authors it outperforms all existing global vision systems with respect to measurement precision and accuracy, high speed and real time operation and reliable tracking of large (theoretically unlimited) number of robots under light intensity changes. The originality of the proposed system lies mainly in specially designed robot marks and robots' poses measuring directly in Bayer format image delivered by the camera. These two measures enable robust pose estimation of the robots with subpixel precision, while the significant simplification of the image processing algorithms ensures tracking of many robots with very high framerates. With algorithms running on a 3 GHz Athlon 64 processor 65 robots can be tracked at 80 fps. Moreover, in order to perform a thorough analysis of the system performances related to defined requirements, we propose a new experimental procedure that can serve as a benchmark for evaluation of other systems for the same purpose. (c) 2007 Elsevier B.V. All rights reserved.