The navigation of small drones within an unknown area requires perception and analysis of their surrounding environments, especially in GNSS-denied regions. A way to achieve this is by reconstructing a 3D dense metric map in real time. It allows the drone to localize itself, plan its trajectory and to be aware of potential static and dynamic obstacles. Visual Inertial Odometry (VIO) algorithms focus on robot localization and trajectory, while Simultaneous Localization and Mapping (SLAM) maintains both localization and mapping. These types of algorithms seem to have reached maturity, and now face new challenges related to real applications in robotics. Studies are now towards robustness and efficiency through new sensors and Deep Learning (DL).