A Scalable Tracking System Using Sparse Dot-Based Markers

Localization in large-scale environment is fundamental for high-degree-of-freedom augmented reality (AR) and virtual reality (VR) applications. However, achieving stable and cost-effective widearea tracking remains challenging, as existing methods often increase system cost and complexity when scaled to large spaces. This paper proposes a vision-based tracking system that encodes spatial relationships between markers to generate a planar layout. We propose a BiLSTM-based network for efficient marker recognition and introduce a robust rematching algorithm to enhance recognition accuracy. The proposed approach scales effectively with sparse markers while maintaining a high recognition rate.