Visual inertial SLAM dense mapping for indoor autonomous navigation

Indoor navigation is a really important topic for many robotics application. The vast majority of tasks that a robot can complete in an indoor environment depends on the fact that the robot is able to orient itself inside it. SLAM algorithms give this ability to the robot but their performance strongly depends on the sensors on which the system is built. This thesis will focus on building a comprehensive starting guide on visual inertial SLAM (in particular ORB-SLAM3), which use only monocular camera to recover camera position and build a map of the surroundings, and accelerometer data to recover real world scale. In this work ORB-SLAM3 will be integrated with a relative depth estimation neural network to build a dense map of the environment rather than a sparse one. Dense map are in fact really fundamental for indoor navigation.