Vision-based　simultaneous　localization　and　mapping　(VSLAM)　which　uses　visual　sensor　to　make　a　robot　locate　itself　in　an　unknown　environment　while　simultaneously　construct　a　map　of　the　environment.　With　the　continuous　development　of　computer　vision　and　robotics,　VSLAM　has　become　a　supporting　technology　for　popular　fields　such　as　unmanned　aerial　vehicle,　virtual　reality　and　unmanned　driving.　In　this　paper,　the　classical　framework　of　visual　SLAM　is　introduced　briefly.　On　this　basis,　the　key　technologies　and　latest　research　progress　of　VSLAM　from　indirect　and　direct　methods　are　surveyed.　Then　the　research　progress　of　deep　learning　techniques　applied　to　VSLAM　is　reviewed.　Finally,　the　development　tendency　of　VSLAM　is　discussed.