Title: Calibration-free probabilistic image sensor model for Bayesian search and tracking

Authors: Shen Hin Lim, Tomonari Furukawa

Addresses: School of Mechanical and Manufacturing Engineering, University of New South Wales, Sydney 2052, New South Wales, Australia. ' School of Mechanical and Manufacturing Engineering, University of New South Wales, Sydney 2052, New South Wales, Australia

Abstract: This paper presents a calibration-free approach to modelling an image sensor in probabilistic manner for recursive Bayesian search and tracking (SaT), such that its information can be utilised interchangeably between |search| and |track| mode. A technique is developed to modify uncertainties from previously established deterministic model into probability density function (PDF). In this technique, the sensor parameters are redefined in probabilistic manner, and, are combined to achieve the resultant sensor model. This model is independent of image data, and hence, is able to provide consistent and reliable PDF of the target state in |search| and |track| modes. Using area coverage method and one-step look ahead as control strategies, the proposed modelling technique was compared with a conventional technique. The model derived from proposed technique provided better estimation of target state and comprehended limitations of image recognition system. This model, in comparison to a pre-calibrated model example, was shown to have higher efficiency and reliability.

Keywords: sensor modelling; image sensors; probabilistic models; recursive Bayesian search; tracking; mechanistic deconvolution; uncertainty; probability density function; PDF.

DOI: 10.1504/IJAAC.2008.022177

International Journal of Automation and Control, 2008 Vol.2 No.2/3, pp.195 - 212

Published online: 22 Dec 2008 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article