Map Learning and Real-time Vehicle Localization for Visual Navigation

In this paper a real-time data fusion approach for vehicle localization to adaptive stabilization of uncertain GPS localization system is presented. Our approach is based on the fusion of GPS, 3D gyroscope and speedometer. The global probability density function (PDF) is adopted to be the blending factor instead of general Kalman gain function, which allows our approach to be robust and accurate for most of practical systematic problems for vehicle localization like slow data drift, large infrequent data jumps. Combining with vision sensor for lane shape recognition and tracking, our system provides a very accurate and real-time vehicle localization approach, which has been adopted to superimpose virtual navigation indicators and icons onto real driver´s view to direct visual navigation in VICNAS^[1] system. Simulation and real road tests verified the effectiveness and efficiency of our approach.