IJCAS

This paper proposes a method for localization of vehicle using one point plus an edge matching region of monocular vision in wide urban environments. The five degree of freedom (5-DoF) localization es-timated by monocular omnidirectional camera improves the planar motion assumption in most of conventional researches. In recent year, the car-like motion model with planar motion is often investigated to reduce the requirements of correspondence until one point. However, in the real appli-cation of long-range motion in outdoor scene, the motion may not satisfy this condition. This leads to the inaccurate vehicle localization. In this proposed method, the car-like model is also utilized for 5-DoF localization however the requirements of correspondence are reduced to only one point plus an edge matching region which is much simpler than the conventional 5-point RANSAC. The cumulative errors of visual odometry are excluded by using global positioning system (GPS) correction under equation of maximum likelihood estimation in Extended Kalman Filter (EKF) frame work. The real application in hills and mountainous regions demonstrates the accuracy of vehicle localization using this proposed method.

Keywords: Chamfer matching, correspondence, EKF, five-point RANSAC, GPS, omni-directional camera, one point RANSAC, visual odometer.