Global mapping and efficient localization algorithm based on ground texture

Abstract

With technological advancements, the deployment of mobile robots in commercial and institutional settings has witnessed a steady rise, significantly contributing to improvements in operational efficiency and service quality. Given the stability of indoor environments, the ground typically exhibits minimal variations and a relatively flat surface, offering robust planar prior information for robotic systems. This paper proposes a systematic approach for mapping and localization by leveraging ground texture captured through a camera, encompassing feature processing, map construction, and rapid localization. An attention-based network is used to extract and match image features, enabling accurate pose estimation and improving visual tracking stability. Furthermore, this paper explores a methodology for constructing a globally consistent map based on ground texture, where the resulting database provides accurate prior information and loop closure frames to support the system’s positioning and tracking phases. Finally, experimental evaluations were conducted across diverse scenarios to assess feature matching rates, loop closure detection, and global localization performance. The results demonstrate a low mismatching rate of approximately 4% in feature matching, alongside high-precision positioning with minimal error, thereby validating the accuracy and robustness of the proposed system.