|M.Sc Student||Rybnikov Silvina|
|Subject||Building a Non-Euclidean Roadmap from a Small Set of Images|
|Department||Department of Computer Science||Supervisors||Professor Ilan Shimshoni|
|Professor Ehud Rivlin|
This work addresses the problem of robotic navigation using natural visual features.
The proposed algorithm receives a set of target locations represented by images taken from those locations. The robot autonomously explores the environment, locates the targets, and builds a graph of the paths between them. The graph allows fast homing from any location in the environment to any target location.
Our approach is unique in that the exploration is autonomous. The problem of finding the locations from which the images were taken is translated into a coverage problem with target identification. The target image is identified, when images overlap, using epipolar geometry estimation techniques. Thus, the camera and the image matching technique can be considered a complex type of sensor. The robot searches for targets by taking images in poses which best improve coverage. We have also developed techniques for improving the pose estimation by using the information from the image matching process.
We present experimental results for a mobile robot in an environment without artificial landmarks. The target locations are far apart and cannot be matched directly. After the graph is built automatically, the robot is placed in an arbitrary position and uses the graph to navigate to a specified location.