This paper addresses the vision-based self-localization of an indoor service robot worked with a single vehicular camera. Based on the principle of projective geometry and the extended virtual image plane, the transverse position and orientation of robot is calculated through considering the image plane and floor plane as a whole. The longitudinal position is acquired by using the monocular stereovision method based on distance constraint along feature points of nature or artificial landmark. After a simple off-line process about landmark, the service robot can work well in a building. The validity and feasibility of the approach have been demonstrated through experiments.