A Framework for Visual Position Estimation for Motor Vehicles

Abstract

This paper describes a general formulation for vehicle position estimation within a road network using visual features from a camera system and a priori knowledge in the form of Geographic Information System (GIS) data. The proposed approach consists of two parts. First, features of the environment are detected by the vision system while corresponding features are extracted from the GIS, which can be considered a system's internal model of the environment. Second, vehicle position is tracked over time using an extended Kalman filtering (EKF) scheme in which visual feature estimates are compared to features extracted from the GIS world model. Simulation results provide a visual illustration of the theoretical finding that uncertainty in vehicle position is reduced by the observation of features changing continuously with vehicle position. This work is applicable for autonomous navigation systems (which must observe the vehicle environment) and as a complement to satellite positioning methods.