Skymask Matching Aided Positioning Using Sky-Pointing Fisheye Camera and 3D City Models in Urban Canyons

Area of Science:

• Computer Vision
• Geomatics Engineering
• Satellite Navigation

Background:

• Dense urban environments pose significant challenges for traditional Global Navigation Satellite System (GNSS) positioning due to signal obstruction and multipath effects.
• Existing 3D-mapping-aided (3DMA) GNSS methods often rely on complex simulations for signal correction, limiting their efficiency.
• There is a need for robust positioning algorithms that can leverage environmental features for improved accuracy in challenging GNSS reception areas.

Discussion:

• This paper presents a novel algorithm utilizing building boundaries for positioning and heading estimation, bypassing complex signal reflection simulations.
• A convolutional neural network is employed to segment fisheye images, distinguishing between sky and building elements.
• A skymask matching algorithm correlates segmented images with a 3D building model to determine precise location and heading.

Key Insights:

• The proposed method achieves degree-level heading accuracy, significantly enhancing orientation estimation.
• Positioning accuracy comparable to advanced 3DMA GNSS methods is demonstrated in complex urban settings.
• The algorithm effectively utilizes building outlines from fisheye imagery for navigation, offering a practical solution for urban GNSS challenges.

Outlook:

• Further research could explore integration with other sensor data for even greater robustness.
• The algorithm's scalability to different urban densities and building types warrants investigation.
• Potential applications include autonomous vehicle navigation, drone positioning, and augmented reality systems in cities.