|M.Sc Student||Timnah Rotem|
|Subject||Simultaneous Localization and Mapping from Aerial|
|Department||Department of Civil and Environmental Engineering||Supervisors||Professor Gilad Even-Tzur|
|Mr. Garry-Haim Zalmanson|
|Full Thesis text - in Hebrew|
Simultaneous Localization and Mapping (SLAM) field originate from Robotic Engineering. It utilizes the sensors onboard the platform to determine the robot location and orientation (i.e. Localization) and mapping the surroundings at the same time. As the robot advances, it acquires new data from the onboard sensors and calculates both the new orientation and the new areas.
Aerial imagery has specific characteristics, as opposed to ground navigation using close range imagery, the long distance to the photographed objects cause a small parallax between the images. The small parallax makes it difficult to solve the camera's parameters and to determine the coordinates of the objects in the world.
The research goal is to produce accurate mapping data using photogrammetry to perform autonomous navigation, without the need of post-processing the data. This objective requires high accuracy and fast processing speed, and thus presents a significant challenge in the photogrammetric research.
This research carried out using a flight simulator of an airborne camera. The flight simulator outputs are the images' coordinates and orientation, and the objects in the simulated world. The data is then processed using progressive Kalman Filter and Bundle Adjustment. The results are then compared to the ground truth and it's analysis show the optimal flight configuration. Additionally, an examination was carried out in order to understand the differences between the progressive solution and the full Bundle Adjustment solution.
From these tests we were able to determine that the GPS accuracy and the geometric relationship between images are highly important factors influencing the overall success of the SLAM algorithm.
By using the optimal flight parameters, we suggested a flight outline that presented continues ground coverage and produce high accuracies. The high accuracy enables the production of mapping products as ortophoto, digital elevation models and 3d visualization.
The last sections of this study are dedicated to a field test that was carried out from the "Giborim" bridge in Haifa. The test was conducted using a simple digital video camera and a bi-frequency geodetic GPS. The field test confirmed the results which were obtained in the simulation section.