Bilkent University
Department of Computer Engineering


Detecting Trees as Landmarks for Outdoor SLAM


Tugba Yildiz
Computer Engineering Department
Bilkent University

The purpose of this study is to identify stable natural objects which are used as landmarks for Simultaneous Localization and Mapping (SLAM) of robots. SLAM of robots involves building a map of the robots environment and localizing the robot with respect to that map simultaneously. In SLAM, the robot needs to recognize landmarks for the sake of using them while building its map. In this study, trees, which can be abundant in an outdoor environment, are detected for the purpose of using them as landmarks for SLAM. However, detection of trees is a difficult problem because trees are very similar to background foliage in terms of color and texture. We describe a computer vision system for detecting trees in outdoor cluttered environments based on the vertical nature of the tree trunks. In the tree trunk detection algorithm, firstly edges dominant in the vertical direction are obtained by using the edge flow segmentation algorithm, then the dominant edges are perceptually organized into continuous and symmetric lines, and lastly, these lines are grouped into potential trees -landmarks- by minimizing the entropy of the line pixels inside the image. The performance of this system will be evaluated on a database of images containing trees.


DATE: 27 April, 2009, Monday@ 17:00