Fisheye-Based Method for GPS Localization Improvement in Unknown Semi-Obstructed Areas

Area of Science:

• Robotics and Autonomous Systems
• Geomatics Engineering
• Computer Vision

Background:

• Precise Global Navigation Satellite System (GNSS) localization is critical for autonomous road vehicles.
• Urban environments with occluded satellites hinder accurate positioning.

Purpose of the Study:

• To fuse Global Positioning System (GPS) data with fisheye stereovision for robust vehicle localization.
• To develop a method for Non Line Of Sight (NLOS) detection and pseudorange delay estimation.

Main Methods:

• Utilized a sky-facing stereoscope with 360° fisheye cameras for 3D modeling and plane extraction.
• Implemented stereoscope self-calibration, stereo matching with epipolar curves, and robust plane fitting.
• Fused 3D data with GPS raw data to estimate NLOS reflected signal pseudorange delay using a 3D canyon model.

Main Results:

• Generated 3D models achieved metric accuracy.
• Demonstrated over 50% improvement in horizontal GPS positioning accuracy.
• The proposed NLOS detection method outperformed Carrier-to-receiver Noise density (CN0)-based methods.

Conclusions:

• The proposed fusion of GPS and fisheye stereovision effectively enhances autonomous vehicle localization.
• The developed NLOS detection technique provides a significant improvement over existing methods.