2014/11/05 by Nima Keivan, Keivan, Nima, Gabe Sibley +1
Engineering · Computer Science · #Robotics and Sensor-Based Localization #Advanced Vision and Imaging #Advanced Image and Video Retrieval Techniques
paper · pdf · doi:10.48550/arxiv.1411.1372
A framework for online simultaneous localization, mapping and\nself-calibration is presented which can detect and handle significant change in\nthe calibration parameters. Estimates are computed in constant-time by\nfactoring the problem and focusing on segments of the trajectory that are most\ninformative for the purposes of calibration. A novel technique is presented to\ndetect the probability that a significant change is present in the calibration\nparameters. The system is then able to re-calibrate. Maximum likelihood\ntrajectory and map estimates are computed using an asynchronous and adaptive\noptimization. The system requires no prior information and is able to\ninitialize without any special motions or routines, or in the case where\nobservability over calibration parameters is delayed. The system is\nexperimentally validated to calibrate camera intrinsic parameters for a\nnonlinear camera model on a monocular dataset featuring a significant zoom\nevent partway through, and achieves high accuracy despite unknown initial\ncalibration parameters. Self-calibration and re-calibration parameters are\nshown to closely match estimates computed using a calibration target. The\naccuracy of the system is demonstrated with SLAM results that achieve sub-1%\ndistance-travel error even in the presence of significant re-calibration\nevents.\n