Abstract: The sparse and dense approaches are two main aspects in vision-based simultaneous localization and mapping (VSLAM). Key technologies and latest research progress of the both approaches are reviewed in detail, and various aspects of the comparative advantages and disadvantages and the implementation difficulties of different methods are discussed. The research progress of deep learning techniques applied to VSLAM is reviewed, and the combination manner of the two approaches for improving performances is discussed. Finally, the future research directions of real-time VSLAM are explored.