Abstract: Inspired by orientation sensation of pigeon,we divide the self-localization approach of mobile robot with omni-vision and odometer into two modes: mode based on omni-vision and mode based on odometer.The robot selects its concrete self-localization mode at certain time according to a certain rule.Firstly,the robot attempts to use the omni-vision information to localize itself.If there is not enough vision information to localize itself,or the reliability of the omni-vision based self-localization cannot be accepted,the robot uses the odometer information to localize itself.For the problem of landmark information distortion,it calculates the theoretic values of landmarks based on the initial visual self-localization results,and compares the theoretic values with the possible values of landmarks extracted from the raw images to filter out the unreliable vision-based self-localization.To gain exact self-localization of moving robot,factors of time delay that affect the precision of self-localization is analyzed.