Monocular visual-inertial odometry system

A real-time Monocular Visual-Inertial State Estimator (VINS-Mono) is used in this project to replace Global Positional System (GPS) in a confined and indoor environment, where GPS measurements are either unavailable or unacceptably inaccurate, to provide environment information for precise state estimation. This environment information is supported by the monocular Visual-Inertial System (VINS) to carry out Visual-Inertial Odometry (VIO). However, the VINS-Mono is very susceptible to long term drifting in global 3- Dimensional (3D) translation and orientation due to the loss of visual tracking. Therefore, the objective of this project was to investigate on the issue of the loss of visual tracking due to pedestrian observed from a stationary monocular VINS that caused the state estimator to fail and thus, improvise on the existing VINS-Mono algorithm to cope with this problem. The aim was then to experiment with key linear acceleration values retrieved from the IMU to minimize or eliminate the uncontrollable drifting when the visual tracking is lost and hence, enhance the stability and reliability of existing VINS-Mono in the light of practical visual disturbances. With the improvised VINS-Mono, the state estimator was able to effectively predict the instance before the drift occurs and prevent the drift from occurring. Consequently, drifts in global 3D translation and orientation was controllable and minimized when compared to the original VINS-Mono. These improvements have a great significance for the practical applications of an autonomous UAV, especially for navigation tasks in a restricted and foreign environment. At the end of this report, I will evaluate on the methods that I had experimented on and suggest recommendations for future work.Bachelor of Engineerin