Extracting and Matching Lines of Low-Textured Region in Close-Range Navigation for Tethered Space Robot

Abstract

When dealing with lines in regions with sparse texture, such as satellite's brackets, some existing line matching methods do not work well due to the incorrect location of line endpoints and line fragments. In this paper, we study how to automatically match low-textured lines. The designed feature only uses their neighborhood appearance and there are no any other prior knowledge or constraints needed. We combine point and line features to propose a novel line matching method. It includes the following three main steps. First, line extraction based on pixel gradient is adopted and we design a mergence strategy to ensure continuity. Then, line-point invariant and center-symmetric local binary pattern descriptor are combined together to represent lines. Last, two corresponding criterions are designed to measure the similarities between each pair images. Extensive experiments on real and synthetic images show that our proposed method exceeds the reference methods in performance under scale, illumination, and dynamic cases.