A road recognition system incorporating GPS/GIS and vision components

This dissertation presents a road recognition system that uses two different components, a Global Positioning System (GPS)/Geographical Information System (GIS) integrated system and a vision system. The desired outputs of the recognition system are the distance of vehicle to the vanishing point and the elevation angle between these two points. The GPS/GIS integrated component makes use of GPS data and Digital Elevation Model (DEM) data from a GIS. The system is capable of determining the geometry of the road ahead with various vehicle speeds and outdoor road conditions. The system is divided into several components including location systems, a reckoning system, map matching, and map database. The system transforms GPS data to map coordinate data. The elevation information corresponding to the current vehicle location is obtained by map matching. The reckoning system computes the desired output using the visibility in terrain method which is used to determine the mutual visibility from the vehicle to the vanishing point in front of the vehicle.; The second system is the vision system based on image processing algorithms. The 3-D reconstruction algorithm relies heavily on several parameters. First, the camera system has to be calibrated to produce accurate intrinsic parameters which will be used to link between a world and camera reference. Second, the edge extraction and matching process algorithms are required to provide precise pixel location and the tangent of road boundary. The 3-D reconstruction algorithm is able to compute a 3-D road geometry according to data obtained from the edge extraction and matching process.; The experiments and results are described to verify the performance and accuracy of each algorithm. An analysis of various vehicle speeds and outdoor conditions has been tested on the GPS/GIS integrated system. In addition, the vision system including camera calibration algorithm, edge extraction and matching process algorithm, and 3-D reconstruction algorithm are tested on different road condition images. Tradeoffs and drawbacks between the two different components are described.