GPS-supported visual SLAM with a rigorous sensor model for a panoramic camera in outdoor environments.

Area of Science:

• Robotics
• Computer Vision
• Geomatics

Background:

• Accurate localization of moving sensors is critical for applications like robot navigation and urban mapping.
• Existing Simultaneous Localization and Mapping (SLAM) methods often rely on ideal sensor models, which can introduce system errors.
• The integration of GPS data and advanced sensor modeling is needed to improve SLAM accuracy and robustness.

Purpose of the Study:

• To develop a GPS-supported visual SLAM framework utilizing a rigorous sensor model for panoramic cameras.
• To address the limitations of ideal sensor models in panoramic camera-based SLAM.
• To enhance global optimization and georeferencing capabilities through GPS integration.

Main Methods:

• A rigorous sensor model for panoramic cameras was established, analyzing and overcoming issues with ideal models.
• GPS data was incorporated for global optimization and georeferencing.
• A GPS-supported Bundle Adjustment-SLAM (BA-SLAM) approach was developed, combining ray and GPS observations.

Main Results:

• The proposed GPS-supported BA-SLAM framework demonstrated high localization accuracy.
• Vehicle-borne panoramic images from a campus environment were used for validation.
• Ground control points confirmed the method's ability to achieve centimeter-level accuracy.

Conclusions:

• The developed framework offers a significant improvement over traditional SLAM methods by employing a rigorous sensor model.
• GPS integration enhances the global accuracy and georeferencing of the SLAM system.
• The method provides a robust and accurate solution for real-world applications such as robot navigation and urban mapping.