SPS: Accurate and Real-Time Semantic Positioning System Based on Low-Cost DEM Maps

Area of Science:

• Computer Vision
• Geographic Information Systems (GIS)
• Robotics

Background:

• Traditional Global Positioning System (GPS) offers limited accuracy for applications like Augmented Reality (AR).
• Outdoor urban environments pose challenges for precise mobile device geo-localization due to GPS signal limitations and lack of detailed map data.

Purpose of the Study:

• To develop a Semantic Positioning System (SPS) for enhancing mobile device geo-localization accuracy in outdoor urban settings.
• To enable precise 6 Degree-of-Freedom (DoF) pose estimation by integrating diverse data sources.

Main Methods:

• Integration of Geographic Information System (GIS) data, GPS signals, and visual image information.
• Cross-view semantic matching using Digital Elevation Model (DEM) data and sampled observation pixels with predicted semantic labels.
• Iterative homography estimation with semantic correspondences and heuristic search for efficient matching.

Main Results:

• Achieved 73.24% positioning accuracy on the Bund dataset, outperforming a baseline method by 20%.
• Improved positioning accuracy by an average of 5% compared to state-of-the-art semantic-based approaches on the Kitti dataset.
• Demonstrated a robust, real-time, and automatic positioning system with excellent scalability.

Conclusions:

• The proposed Semantic Positioning System (SPS) significantly enhances geo-localization accuracy for mobile devices in urban environments.
• The system effectively overcomes limitations of DEM data by utilizing semantic information for precise pose estimation.
• SPS provides a scalable and efficient solution for AR and other location-aware applications.