Using match uncertainty in the Kalman filter for a sonar based positioning system

Abstract

Much effort has been devoted to computer vision methods for navigation and tracking of flights, cars, robots, and autonomous vehicles. Less work has addressed the problems of visual navigation under water, which is the topic of this paper. We present a method for incorporating the uncertainty of the matching between a model of the sea floor and measurements, obtained with a multibeam sonar, in a Kalman filter. The technique is based on using second order centralised moments in a region of interest to estimate the measurement uncertainty/error. This error is then used to calculate the measurement covariance matrix in the Kalman filter. We provide experimental results on real data to show that this method is superior to the standard Kalman filter in which the measurement error covariance matrix typically is set constant or varies deterministically. 1. Introduction Navigation has always been an interesting area for researchers. Recently, there has been considerable interest in using com...