In　this　paper,　a　micro　SLAM　system　based　on　ORB　features　for　RGB-D　cameras　has　been　proposed.　With　only　a　RGB-D　sensor,　this　method　can　be　applied　in　small　environment　for　localization　and　mapping.　Furthermore,　the　task　of　3D　reconstruction　can　also　be　accomplished　by　using　the　approach.　The　pose　graph　based　on　Bundle　Adjustment　is　adopted　for　reducing　the　estimation　error.　In　order　to　further　speed　up　computing　to　meet　the　requirement　of　real-time,　we　have　proposed　the　piecewise　optimization　strategy.　The　approach　is　evaluated　on　public　benchmark　datasets.　Compared　with　several　state-of-the-art　scheme,　this　method　has　proven　to　work　well　in　these　environments.　©　The　Authors,　published　by　EDP　Sciences,　2018.