In　recent　years,　much　research　has　been　done　on　RGB-D　simultaneous　localization　and　mapping　(SLAM)　system.　Mismatching　and　depth　measurement　uncertainty　are　key　factors　affecting　the　accuracy　of　RGB-D　SLAM　algorithms.　Given　that,　we　propose　a　3D　point　correspondences　uncertainty　aware　SLAM　system.　Firstly,　we　conduct　ORB　feature　extraction　and　matching.　Secondly.　3D　positions　of　those　pair　points　are　reconstructed　by　combing　depth　information　with　Gaussian　Mixture　Model　(GMM)　and　outliers　are　rejected　and　the　initial　guess　of　camera　motion　can　be　provided　based　on　LMedS.　Under　the　assumption　of　Chi-Square　distribution,　the　motion　results　are　further　optimized　by　using　Mahalanobis　distance　and　Chi-Square　test.　Besides,　the　camera　motion　trajectory　is　globally　optimized　by　pose　graph.　Finally,　experiment　results　prove　the　proposed　method　can　improve　the　accuracy　of　localization　and　mapping.