Recovering structure uncertainties from noisy sense data Tarek M. Sobh * , Ausif Mahmood Department of Computer Science and Engineering, University of Bridgeport, 169 University Avenue, Bridgeport, CT 06601, USA Received 29 June 1998; accepted 28 March 2000 Abstract This work examines closely the possibilities for errors, mistakes and uncertainties in sensing systems. We identify and suggest techniques for modeling, analyzing, and recovering these uncertainties. This work concentrates on uncertainties in visual sensing for manipulators. The goal is to recover three-dimensional structure and motion characteristics of the environments under observation from noisy measurements. We also conjecture that the approaches described here are suitable for other sensors and parameters to be recovered. The computed uncertainties are utilized for reconstructing the geometry, motion parameters, and structure parameters under observation. Ó 2002 Published by Elsevier Science Ltd. Keywords: Sensing, computer vision; Structure from vision; Noisy sensory data; Recovering depth parameters from vision 1. Introduction In this work we discuss uncertainty modeling for sensor systems. In particular, we describe some techniques for measuring and computing the uncertainties in recovering some visual pa- rameters. We concentrate on presenting the sources of uncertainty in two-dimensional 2-D) visual data. Then we proceed to identify methods by which the 2-D uncertainty could be transformed into meaningful three-dimensional 3-D) interpretations that the observer can use reliably in order to recover the world events. Those methods can be generalized for other sensing problems and parametric recovery from sense data. Fig. 1 depicts the sequence of steps that are to be performed in order to recover the full world uncertainty from 2-D measurements on the image plane. We start by recognizing the sensor Computers and Electrical Engineering 28 2002) 127±141 www.elsevier.com/locate/compeleceng * Corresponding author. Tel.: +1-203-576-4116; fax: +1-203-576-4766. E-mail address: sobh@bridgeport.edu T.M. Sobh). 0045-7906/02/$ - see front matter Ó 2002 Published by Elsevier Science Ltd. PII:S0045-790600)00019-7