A　mobile　robot　localization　method　based　on　the　monocular　vision　is　proposed.　According　to　the　planar　ground　assumption,　a　plane　homography-based　mobile　robot　pose　estimation　model　is　established　with　the　KLT　corner　detection　and　tracking　algorithm.　Under　the　framework　of　random　sample　consensus　algorithm,　the　initial　location　of　the　mobile　robot　is　realized　by　solving　the　above　model.　On　this　basis,　a　triangular　structure　of　corner　features　is　adopted　for　plane　parameter　estimation.　Combined　with　M-estimation　algorithm,　the　non-planar　feature　information　is　excluded　efficiently　and　the　location　accuracy　of　mobile　robot　is　improved.　The　indoor　experimental　results　demonstrate　the　effectiveness　and　feasibility　of　the　proposed　algorithm.　©,　2015,　Chinese　Optical　Society.　All　right　reserved.