Fusion and uncertainty modelling of 3-D multisensory data

Dealing with uncertainty for data extracted from disparate sensors is a fundamental issue in the area of sensor integration and fusion. In this thesis we describe a technique that models and fuses three dimensional information extracted from a stereo vision system and an ultrasonic range system in order to recover the three dimensional pose of an ol:)ject as well as the accuracy associated with these measurement. Specifically, we have probabilistically modelled the sources of error in a stereo system and those inherent in the ultrasonic range measurement system. First the measurement generated by stereo system is updated by the ultrasonic system using an optimization technique solved for by the Euler-Lagrange Calculus of Variations equations. Second the uncertainties inherent in each of the measurements are combined by fusing the probability densities of the stereo and ultrasonic range data for each of the target points describing the pose. The advantage of this method is that we are able to deal with significantly different data in a common and unique way to infer knowledge that exploits the best features and ignores weaknesses of each sensor.