Ground Feature Point Matching Based Global Localization for Driverless Vehicles

Vehicle localization is achieved by a ground feature points based map matching approach,in which a camera is fixed downward on the bottom of the vehicle according to the outdoor environmental conditions.The proposed approach includes two steps:(1) a vehicle is manually controlled to move in an environment,recording sensor data from RTK(real-time kinematic)-GPS,odometry and camera to produce a ground feature point map automatically in an off-line manner.A special map organization is used to increase the efficiency of map search and matching.(2) vehicle localization is realized by map matching method,in which a M-estimator weighted ICP(iterative closest point) algorithm is utilized to match feature points and compute matching parameters.Furthermore,map matching result is fused with dead-reckoning by UKF(unscented Kalman filter) to achieve higher robustness.Experimental results demonstrate the effectiveness of the proposed approach.