|Ph.D Student||Indelman Vadim|
|Subject||Navigation Performance Enhancement Using Online Mosaicking|
|Department||Department of Aerospace Engineering||Supervisors||Professor Pinchas Gurfil|
|Professor Ehud Rivlin|
|Dr. Hector Rotstein|
|Full Thesis text|
This research provides a few contributions to the vision-aided navigation literature. The first part of this dissertation focuses on vision-aided navigation of a single platform, which is assumed to be equipped with a standard inertial navigation system and a single camera only.
In the first algorithm, developed in this research, the main idea is to couple online mosaic construction process to a camera scanning pattern, assuming that the camera is mounted on gimbals. It is shown that improved vision-based motion estimation and navigation performance is obtained in challenging operational scenarios.
The second algorithm utilizes constraints, obtained by observing the same scene from three different views, for navigation aiding. A new formulation of such constraints is presented and proven, and a Kalman filter formulation is developed for fusing these three-view constraints with a standard inertial navigation system. Given three images with a common overlapping area, the new algorithm reduces the position errors in all axes to the level of errors present while the first two images were captured. Errors in other navigation parameters are also reduced. Trajectories that contain loops, in which the platform revisits a scene after some unknown time, are naturally handled by the new algorithm.
The second part of this research is concerned with cooperative navigation. A general multi-platform measurement model is considered. This measurement model involves navigation data and readings of onboard sensors from different platforms, possibly taken at different time instances. Since, in the general case, these various sources of information are correlated, the appropriate correlation terms must be calculated to obtain a consistent state estimation. A new method was developed for on-demand calculation of the required correlation terms based on the history of all the multi-platform measurements performed thus far. The newly-developed method relies on graph theory and is capable of rigorously handling the involved process and measurement noise for general multi-platform measurement models.
Finally, a new approach for vision-aided cooperative navigation was developed. In this approach, a measurement is formulated whenever the same scene is observed by three views, possibly captured by different platforms, not necessarily at the same time. As in case of a single platform, applying the three-view constraints for cooperative navigation reduces the position and velocity errors in all axes, as well as other navigation errors, without utilizing range measurements. In the proposed approach the platforms' cameras are not required to be aimed at other platforms.