This dissertation considers the problems of map-aided vehicle localization and map-based road curvature estimation through the use of sensor fusion and nonlinear filtering techniques. The proposed methods are based on the novel concept, Digital Maps as Virtual Sensors, and differ from conventional approaches in terms of the usage of digital map data and estimation methodologies. Conventionally, the use of digital map data for automotive applications is nothing more than data mining and association based on the position information provided by on-board sensors, such as Global Positioning System (GPS) and inertial navigation systems. Thus, most researches are focused on the development of map data mining and association techniques. This dissertation is not concerned with these two techniques, but instead with novel sensor fusion and nonlinear filtering techniques. The goal is to use digital maps as virtual sensors to improve or enable automotive active safety systems. The novelty in terms of methodology for state and parameter estimation is described below. For the vehicle localization problem, the correlations between vehicle states (i.e., position and heading), and the parameters of associated roads (e.g., road heading, number of lanes, and lane/road width, etc.), are taken into account in designing the map-aided vehicle positioning system. A sensor fusion framework for integrating GPS, dead reckoning sensors and the virtual sensor, namely, digital maps is presented in this dissertation. Based on this framework, two map-aided vehicle localization systems are developed: (1) in the first approach, the extended Kalman Filter estimate is conditionally corrected through the use of map data by Map Matching; (2) in the second approach, map information is formulated as state constraints in the constrained unscented Kalman Filter algorithm. Both systems can achieve the desired lane-level positioning accuracy under GPS-degraded conditions in field tests. For the road curvature estimation problem, a dynamic road curve reconstruction method is developed to predict an upcoming curve and select curve points based on digital map data. Through parametric curve fitting, accurate road curvatures can be estimated in real-time. This proposed method is superior to the conventional approach using splines because the resulting curvature derived from splines is very sensitive to map data positional errors, which can not be easily distinguished from the possible detailed road features in real-time. In summary, this dissertation illustrates the novel usage of digital map data for vehicle state and road parameter estimation. The theory presented in this dissertation has been successfully applied to the development of the map-based driver safety assistance system at California PATH Program.