Design, implementation and deployment of a visual odometry system based on a stereovision module

This thesis presents the development of a stereo visual odometry system designed as part of the Nautilus student project, aimed at supporting navigation tasks in challenging environments. Visual odometry allows estimating the motion of a camera-equipped system based solely on visual information, making it particularly valuable in scenarios where GPS or other localization methods are unavailable. The system was tested on three different datasets: the KITTI benchmark, a custom dataset acquired in a controlled environment, and a third collected along the canals of Chioggia during a Nautilus field mission. The analysis highlights both the strengths and limitations of the system in varying conditions. The thesis concludes by discussing possible future developments for the improvement of the system in terms of robustness and accuracy, considering possible future underwater environments.