The Taub Faculty of Computer Science Events and Talks

Since the Global Positioning System (GPS) was established in the 1970s, navigation has become a much easier task. Indeed, the majority of navigation systems rely on the GPS signal for correcting the developing dead reckoning errors. However, GPS is unavailable or unreliable indoors, underwater, in urban environments, and on other planets. In these scenarios, one must use alternative techniques for updating the Inertial Navigation System (INS), or any other dead reckoning mechanism being used. This research presents algorithms that use imagery captured during motion and mosaics constructed based on these images, for performing navigation aiding. The vehicle is assumed to be equipped with an INS and a single camera. No other sensors or a-priori information are required.

First, an algorithm for navigation aiding will be presented which couples between the process of online mosaic construction and the process of a gimbaled camera scanning. Improved navigation precision is obtained, compared to a basic method for image-aided navigation, when operating in challenging scenarios that include a narrow field-of-view camera observing low-texture scenes.

Next, a new set of constraints will be derived by analyzing a general static scene, observed from three different views. A formulation of an implicit extended Kalman filter will be presented for fusing these constraints with an INS. The developed method reduces the navigation errors, including position and velocity errors in all axes without explicit scale recovery. In particular, the method efficiently handles scenarios in which the same scene is observed several times.

A natural extension of this method is to consider a group of cooperative platforms that exchange imagery and navigation information. The three-view constraints are formulated whenever the same scene is observed by different platforms, not necessarily at the same time. Since the navigation information involved in the three-view constraints may be statistically dependant, the appropriate correlation terms should be known in order to obtain consistent information fusion. A new graph-theory-based technique will be presented for on-demand calculation of the required correlation terms for a general multi-platform measurement model.

Both simulation and experimental results will be presented validating the algorithm performance.

Ph.D. seminar under the supervision of Prof. Pini Gurfil, faculty of Aerospace Engineering, Prof. Ehud Rivlin, department of Computer Science, and Dr. Hector Rotstein, Rafael.