Software and Hardware for Pattern Recognition and Image Analysis
Cite this article as:
Eggers, M., Dikov, V., Mayer, C. et al. Pattern Recognit. Image Anal. (2013) 23: 481. doi:10.1134/S1054661813040032
This paper describes the setup and realization of a distributed camera system designed to survey a laboratory area where humans and mobile manipulator robots collaborate jointly. The system consists of 40 industrial grade cameras surveying a 10 m by 10 m area from a top-down perspective, connected via Gigabit Ethernet (GigE) to a cluster of 40 computers for distributed image processing. The cameras were fully calibrated, achieving an average reprojection error of 0.13 pixels for the complete system, which exceeds state-of-the art accuracy. Current long-term testing has the system running with at least 99.994% availability for up to two weeks. Successful application tests of the system were conducted, where it was used to track the movements of robots and humans across the surveyed area.
computer vision multi-camera systems multi-camera calibration performance evaluation