MGG: Monocular Global Geolocation for Outdoor Long-Range Targets

Area of Science:

• Computer Vision
• Robotics
• Geospatial Analysis

Background:

• Traditional monocular vision localization is limited to short-range and indoor relative positioning.
• Accurate global geolocation of long-range outdoor targets using monocular vision remains a significant challenge.

Purpose of the Study:

• To present MGG, a novel monocular global geolocation method for outdoor long-range targets.
• To output precise GPS information for targets using a single RGB image and navigation parameters.
• To demonstrate the method's superiority over existing techniques in challenging outdoor environments.

Main Methods:

• Camera pose correction via pixel mapping.
• Anchor-based methods for improved detection of small, long-range targets.
• Local Monocular Vision Model (LMVM) with local structure coefficient for 2D-to-3D mapping.
• Soft Correspondence Constraint (SCC) to decouple detection and localization.
• Optimization theory and coordinate transformations for final geolocation.

Main Results:

• MGG successfully provides GPS information for outdoor targets up to 150 meters away.
• The method demonstrates superior performance on the KITTI dataset.
• Focal length's critical role in mitigating error explosion for long-range monocular localization is highlighted.

Conclusions:

• MGG offers a robust solution for monocular global geolocation of long-range outdoor targets.
• The proposed techniques effectively address challenges in detecting and localizing distant objects.
• This work advances the capabilities of vision-based navigation and positioning systems.