Advanced vehicle location (AVL) needs vector maps. Two key procedures for accurately generating a vector map are (i) automatic extraction of a road network layer from a raster map to a vector map, and (ii) accurate geo-adjusting for calibration of the vector map. Their main corresponding problems are that the road pattern recognition is not measurable and the map geo-adjusting process is not accurate. This paper presents two new approaches to the solutions through (i) a closed-loop optimal control to the road and non-road pattern recognition to automatically adjust their evaluation function threshold and parameter values, and (ii) a topological discrete nonlinear adjusting method to accurately locate a group of selective points and their neighboring areas for the whole map. This approach has been successfully applied to accurate map vectorization and geo-adjusting Source.