Vision Based Global Localization for Intelligent Vehicles

In this paper, we proposed a vision based global localization approach for intelligent vehicles. A single camera is used to determine vehicle's lateral and longitudinal offsets with respect to the road. Since the number of horizontal landmarks on the road is limited, an extended Kalman filter is used to fuse the results of odometry and vision, which also improves the system's reliability in case that landmarks disappear from camera's field of view. If locations of the landmarks are known a priori, the global pose of the vehicle can be estimated by the proposed methods. The algorithm is composed of two steps: landmarks detection using randomized Hough transform and data fusion with odometry. Experimental results with real data prove the high accuracy and low computation