This paper presents a calibration-free approach to modeling an image sensor for deriving observation likelihood in recursive Bayesian estimation. The previously defined deterministic technique of calibration-free image sensor modeling is utilized to evaluate the uncertainties of the image sensor in probability density function (PDF), and then defines the resultant probabilistic sensor model in recursive Bayesian estimation. This model is independent of image data, and hence, is able to provide consistent and reliable PDF of the target state in both detection and non-detection modes. The probabilistic image sensor model derived by the proposed modeling was evaluated for its feasibility in stochastic localization and its reliability in recursive Bayesian search and tracking, which is a scenario that requires both detection and non-detection modes. The model was also evaluated using both static and dynamic target states. This, in comparison to a model derived by conventional technique, was shown to have higher feasibility and reliability, and provided better target estimation state in both stochastic localization and recursive Bayesian search and tracking.
展开▼
