Learning to Predict Lane Occupancy Using GPS and Digital Maps

Current digital maps are inadequate for many advanced information applications in vehicles. One shortcoming is the lack of lane geometry models with suucient precision. This paper discuss several applications of a lane model database and presents an inexpensive method of inducing lane models by analyzing unsupervised traces of vehicle positions that come from Global Positioning System receivers with diierential corrections. The computed lane models enable safety applications, such as lanekeeping, and convenience applications, such as lane-changing advice. Experiments show that, starting from a baseline map that is commercially available, our lane models predict a vehicle's lane with high accuracy from a small number of passes over a particular road segment. We also use the lane models to predict when a vehicle is changing lanes and compare this with ground truth. In addition to lane models, the analysis of position traces provides a cheap, automated method of deriving other road features, like traac signals and elevations, transforming digital maps into a critical resource for navigation and safety.