| Landmark location uncertainty in digital images, which is extensively used in high precision photogrammetry and machine vision applications, consist of the error measurement when locating the position of a specific image feature.; Landmark location uncertainty has been previously described in the literature for particular landmark designs within the scope of specific applications and using simplified models. For the first time, a general framework to determine landmark location uncertainty in presented in this work. The framework includes the determination of the performance floor by means of the Cramer-Rao Lower Bound (CRLB). The methodology presented, considers the complete physical model of image formation, including 6 degree of freedom, landmark to camera geometry, diffraction, defocus, lens distortion, gray-scale, pixel geometry, and pixel sensitive area.; With the framework developed, an analysis tool was created to model true engineering cases to allow the investigator to predict performance for any configuration of landmark, camera, imager and estimator used. This tool includes the determination of the CRLB performance floor for the configuration used.; Additionally, this work also pioneers novel landmark location estimation algorithms with confidence intervals at tens of milli-pixel level, which not only perform more than 10 times better than existing estimation algorithms but also has been experimentally verified.; The Cramer-Rao Lower Bound methodology introduced in the present work establishes a theoretical statistical minimum limit on the landmark location uncertainty. Knowledge of this bound provides the means to evaluate the actual performance of both existing and future landmark location estimators.; The approach presented in this work includes a mix of analysis, where feasible, and numerical work where required, including numerically deriving the partial derivatives needed to compute statistical distributions and the Cramer-Rao Lower Bound. |
