Creating a distortion characterisation dataset for visual band cameras using fiducial markers

Abstract

be This paper presents two datasets for the purpose of calibrating, evaluating the calibration accuracy, and evaluating stereo vision accuracy of a pair of cameras. The authors provide a baseline that can be used as an initial comparison. This will allow other researchers to perform the same steps and create better algorithms to accurately locate fiducial markers and calibrate cameras. A second dataset that can be used to assess the accuracy of the stereo vision of two calibrated cameras is also provided. A fiducial marker was presented to a camera in a repeatable sequence of movements at different roll angles. Using the information captured during this stage it is possible to characterise the lens distortion of the camera. The calibrated cameras were then used with the stereo vision assessment dataset to triangulate the displacement between a pair of fiducial markers. The results show that it is indeed possible to use fiducial markers to calibrate visual band cameras but sub-pixel localisation would improve the results further.