In this paper, stereo visual odometry (VO) aiding to GPS (Global Positioning System) is investigated, which was inspired by the increasingly widely applied vision-based navigation system. As a self-contained system, stereo visual odometry has been validated for short-term navigation, which makes it capable of not only providing the dynamic information of the land vehicles for GPS positioning, but also bridging the outage of GPS in harsh GPS environment. A simplified tightly coupled GPS/Stereo visual odometry system is developed using Extended Kalman filter (EKF) with error-state to fuse two types of datasets. The system model is derived based on the ego-motion estimation of stereo visual odometry, while the measurement update in EKF is provided by the GPS code and carrier-phase observations. In this way, the ego-motion estimation based on images provides the dynamic information of land vehicle, and the absolute positioning information of GPS signals can mitigate the accumulated errors of visual odometry. PPP (Precise Point Positioning) algorithms are applied to improve the performance with a single GPS receiver. Two datasets were collected near the campus of the University of Calgary in open-sky and harsh scenarios with two PixeLINK cameras and a Trimble R8 amounted on a land vehicle. In each dataset, approximate 4000 images are recorded with 10 FPS (frames per second). A base station was also set up on the roof of the Engineering Building at the University of Calgary to perform DGPS (Differential GPS), which is used as trajectory reference. The results show that with full operation of GPS, the GPS measurements dominate the positioning accuracy. In harsh environment, the stereo VO mitigated the outage of GPS with horizontal error within 20 meters over one minute.
