Robust self localization by edge feature tracking

Feature tracking algorithms have traditionally relied on point features and the use of edges as features is under-investigated. Our paper addresses the “Line segment” as the highlighted edge feature. In order to track the visible motion between two images, we have used the intersection points between pairs of line segments, which are more stable than the center or end points or any single line-segment reference. In this paper, we address at length the problems associated with segment extraction and use a method which breaks curved edge into a reliable straight line segment by putting a limit on its curvature. A straightforward approach toward tracking is discussed by defining geometric information as well as neighboring elements for a line segment feature. Finally, we conclude with a technique to reject outlier matches of line segments by using virtual intersection points to improve tracking.