An Imaging Sensor-Aided Vision Navigation Approach that Uses a Geo-Referenced Image Database.

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Geomatics Engineering

Background:

• Vision navigation is crucial for autonomous systems, particularly in GPS-denied environments.
• Existing methods often rely heavily on Global Positioning System (GPS) and Inertial Measurement Units (IMUs), limiting performance in challenging conditions.
• Mobile mapping systems require robust and precise navigation solutions for diverse applications.

Purpose of the Study:

• To develop a novel vision navigation approach utilizing a high-accuracy geo-referenced image database (GRID).
• To enable high-precision navigation for multi-sensor platforms in environments with poor or absent GPS signals.
• To improve the efficiency and robustness of vision-based localization for mobile mapping.

Main Methods:

• A framework for GRID-aided vision navigation was developed using sequential images from multi-sensor land-based mobile mapping systems.
• An efficient GRID storage management model was established using a linear index of road segments for rapid image retrieval.
• A robust image matching algorithm was designed to accurately match real-time scene images with the GRID database.

Main Results:

• The proposed approach demonstrated efficient image retrieval capabilities.
• Navigation accuracies of 1.2 m (plane) and 1.8 m (height) were achieved under simulated GPS loss conditions (5 minutes, 1500 m range).
• The system effectively calculates 3D navigation parameters for multi-sensor platforms.

Conclusions:

• GRID-aided vision navigation offers a viable solution for high-precision localization in GPS-limited scenarios.
• The developed methods provide efficient data management and robust image matching for real-time applications.
• This approach significantly enhances the reliability of mobile mapping systems in challenging navigation environments.