A camera on-line recalibration framework using SIFT

Camera calibration is a necessary step in many computer vision or photogrammetry tasks. During the execution of these tasks, initial camera calibration may be no longer valid because of intentional or accidental changes of camera parameters. We propose a camera on-line recalibration framework which is aimed at automatically maintaining the calibration of computer vision or photogrammetry system without using any particular calibration pattern again and interrupting the execution of the tasks. The proposed framework consists of initial calibration, followed by recalibration based on SIFT feature point detector and descriptor and a feature point match strategy proposed by us. Both synthetic data and real data have been used to test the framework, and very good results have been obtained. The experimental results of the framework are also compared with those of general on-line self-calibration methods. Both accuracy and speed are also reported. The proposed on-line recalibration framework yields higher accuracy and higher speed on calibrating camera parameters than on-line self-calibration methods do.