My bachelor thesis on comparing Monocular Visual-Inertial Odometry methods.

Abstract:
Monocular Visual-Inertial Odometry (VIO) refers to the process of estimating a robot’s position and velocity relative to its environment using a single camera (monocular vision) in combination with inertial sensors. This project focuses specifically on applying and benchmarking VIO methods in aerial platforms — particularly Micro Aerial Vehicles (MAVs) — although many of the insights gained are also applicable to other drone platforms as well as AR/VR systems. We implemented and evaluated multiple state-of-the-art VIO systems on an aerial drone platform. Our findings reveal significant variation in the performance of these methods — both in terms of trajectory accuracy and computational efficiency — highlighting the importance of careful system selection and tuning in real-world deployment scenarios.