FUSION OF GPS AND MACHINE VISION FOR ABSOLUTE VEHICLE POSITIONING

This paper describes the use of data from a Geographical Information System (GIS) as a priori knowledge to be exploited for reliable and consistent vehicle positioning. The line-of-sight approach taken by GPS creates location inaccuracies especially in urban areas, or leads to complete signal loss when underground, in tunnels, etc. Many attempts have been made to complement GPS with dead reckoning sensors for the purpose of more reliable vehicle positioning; however, such sensors typically suffer from accumulated error. Our approach incorporates a vision system to detect road features and landmarks that can be matched to data from a GIS to determine the position of the vehicle. This information can be fused with the GPS position estimate to increase the reliability of the overall position estimate. The GIS knowledge base incorporates several types of information on different layers, such as road shape, location of landmarks such as buildings, and surrounding terrain. Furthermore, the information available from GIS about the environment surrounding the vehicle position allows constraints to be placed on the possible vehicle positions, creating a positioning system that is highly reliable and aware of its environment. An issue with the type of GIS data we have used is the disparity in representation of features versus a visual representation, which we shall discuss herein and provide suggestions on how to address.