Publication: Monocular visual – inertial odometry for 3d positioning
Loading...
Date
2020-07-01
Authors
Choong, Yi Fung
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Odometry is a localization technique that has been widely employed in an autonomous mobile platform. In general, odometry estimates pose of a mobile platform over time and
categorized according to the onboard sensor measurements used in performing localization. It is convinced that the fusing of data from a proprioceptive sensor and an exteroceptive sensor will yield better accuracy in comparison to that of the standalone odometry. Hence, the objective of the research is to study the application of a monocular visual-inertial odometry and compare the respective performance to that of the inertial odometry and the visual odometry. The proposed visual-inertial odometry is to be implemented using MATLAB and simulate on KITTI dataset. The first step is to construct the proposed method of visual-inertial odometry based on the Multi-State Constraint Kalman Filter (MSCKF) algorithm, which include the initialization of the state vector, state propagation, state augmentation, measurement model and the state update. Next, construct the image processing frontend where most of the work are implemented using the predefined functions from MATLAB vison toolbox. For the experiment of the research, the proposed method is tested along with the use of an inertial odometry and a visual odometry, with the ground truth set as the reference path. For each odometry, a traverse path is constructed and analysed using the information obtained from the respective path such as deviations with reference to the ground truth measurements. A root means square error (RMSE) test is conducted to further validate the result where it is inferred that the implementation of sensor fusion will reduce the RMSE yielded by standalone odometry effectively. The parameters of the proposed method are tuned and tested to increase the chance of best-case scenario. In a nutshell, the proposed monocular visual-inertial odometry able to perform localization and proven to be more accurate than that of the inertial odometry and a visual odometry.