Abstract: A method of robot homing based on panoramic vision is proposed.The omni-directional images of home position can be collected through the panoramic vision device.Then,feature points in the omni-directional image are extracted by adopting SURF(Speeded-Up Robust Feature) algorithm and are used as natural landmarks.During the course of homing, the panoramic images of home position and current position are matched using SURF algorithm,so that the corresponding relations of the natural landmarks between the current and the home images are determined.A method for estimating robot rotation angle is proposed,and fault landmark matches can be eliminated.According to changes of the landmarks' included angles between the home position and current position,the home direction can be calculated to control robot homing.When the included angle between landmarks reaches a threshold value,the robot gets to the original orientation and completes homing mission.The experiment results show that the position error is less than 0.05 m,and the azimuth error is less than 3°. This method doesn't use manual landmarks nor angle sensors,and can be applied to service robots,robot soccer and so on.