Landmark tracking is key factor for mobile robots localization and navigation. This paper proposes a combined approach automatically to detect and track landmark. Firstly, a landmark is initially located in the image coordinates by features recognition- SIFT (Scale Invariant Feature Transform) and matching technology-RANSAC(Random Sample Consensus). Then based on similarity distance, tracking algorithm is called, which depends on adaptive particle filter. Furthermore, re-position strategy based SIFT is also presented to catch the landmark which was lost. Finally, the experimental results show that the proposed method achieves robust and real-time tracking of a landmark and has a practical value for robot visual.