In　order　to　solve　the　localization　problem　of　the　simultaneous　localization　and　mapping　(SLAM),　a　method　of　indoor　robot＇s　self-localization　algorithm　based　on　RGB-D　sensor　was　proposed.　Firstly,　the　feature　points　set　were　extracted　from　the　environment　pictures　by　SURF　(Speed　Up　Robust　Features)　method　and　the　corresponding　matching　points　were　found　using　nearest　neighbor　method.　Secondly,　the　matching　points＇　3D　data　were　reconstructed　by　using　a　combination　of　vision　data　and　depth　data.　Thirdly,　the　robot＇s　movement　parameters　which　are　consisted　of　rotation　matrix　and　transport　vector　are　estimated　by　SVD　and　optimized　by　ICP　algorithm.　Finally,　experimental　results　indicated　the　validation　of　the　proposed　method.