Camera　pose　estimation　is　a　key　step　in　SLAM　system,　which　affects　the　accuracy　and　efficiency　of　the　whole　SLAM　system.　At　present,　there　are　two　main　methods　to　estimate　the　pose　of　camera,　namely,　the　feature　point　method　and　the　direct　method.　The　accuracy　of　feature　point　method　depends　on　the　number　of　feature　points　and　the　correctness　of　feature　matching.　When　enough　feature　points　cannot　be　extracted　in　the　scene,　the　position　and　posture　of　the　camera　cannot　be　estimated　accurately.　The　direct　method　estimates　the　position　of　the　camera　by　the　pixel＇s　photometric　error,　and　does　not　need　to　extract　the　feature　points.　Therefore,　the　direct　method　can　still　estimate　the　position　and　pose　of　the　camera　more　accurately　when　the　feature　point　method　is　unable　to　work.　But　the　direct　method　assumes　the　luminosity　invariance,　so　the　accuracy　of　the　result　is　not　as　good　as　that　of　the　characteristic　point　method.　Aiming　at　the　problems　of　camera　pose　estimation　in　SLAM,　an　improved　camera　pose　estimation　method　is　proposed　in　this　paper.　The　main　idea　of　this　method　is　to　combine　the　feature　point　method　with　the　direct　method　to　overcome　the　estimation　of　the　position　and　posture　of　the　camera　when　the　feature　point　is　lacking,　and　to　improve　the　accuracy　and　robustness　of　the　position　and　posture　of　the　camera.　In　particular,　first,　a　feature　matching　algorithm　which　combines　camera　motion　model　with　image　division　is　proposed.　The　algorithm　improves　the　accuracy　and　quantity　of　feature　matching　while　guaranteeing　the　matching　speed.　Secondly,　on　the　basis　of　the　feature　points,　by　introducing　the　photometric　information,　an　apparent　shape　weighted　fusion　method　is　proposed　to　estimate　the　position　and　posture　of　the　camera.　This　method　can　still　work　steadily　when　the　feature　points　are　lacking.　Finally,　on　the　basis　of　the　preferred　key　frame,　the　local　and　global　fusion　of　camera　pose　optimization　is　realized,　in　which　the　local　optimization　is　realized　by　constructing　the　common　view　relationship　of　the　local　key　frame,　and　the　global　optimization　is　realized　by　the　pose　graph　based　on　the　closed　loop　detection.　In　order　to　verify　the　performance　of　the　pose　optimization　method,　a　SLAM　system　based　on　this　method　is　constructed,　and　the　reconstruction　experiments　are　carried　out　on　the　current　popular　scene　image　data　set.　The　reconstruction　results　verify　the　effectiveness　of　this　method.　©　2018,　Editorial　Department,　Journal　of　South　China　University　of　Technology.　All　right　reserved.