Embedded SLAM algorithm based on ORB-SLAM2

The goal of this dissertation was to present a comprehensive analysis of the ORB-SLAM2 algorithm. By introducing the basis of points triangulation and ORB features, the whole structure of the algorithm has been analyzed, with a centralized focus on the graph-based optimization involved, as well as the place recognition mechanism. Additionally, the original code has been analyzed and optimized, resulting in a substantial increase in time performance while keeping a similar accuracy to the original one, proved by several simulations performed. The final goal of this thesis has been the testing of the obtained algorithm on a real quadcopter and the analysis of its outcomes: the results were affected by the limited computational resources available in the vehicle, obtaining a lower accuracy with respect to the simulation, but still proving the efficiency of the improvements applied on the original code.