A Real-Time Registration Algorithm of UAV Aerial Images Based on Feature Matching

Area of Science:

• Geospatial technology
• Computer vision
• Robotics

Background:

• Accurate real-time geographic positioning of Unmanned Aerial Vehicle (UAV) aerial image targets is challenging.
• Existing feature-matching algorithms struggle with UAVs' rapid motion, camera changes, and high-resolution, sparse-feature maps, leading to mismatches.

Purpose of the Study:

• To develop an accurate and real-time method for registering UAV camera images onto geographic maps.
• To overcome limitations of current feature-matching algorithms in dynamic UAV scenarios.

Main Methods:

• Employed the SuperGlue algorithm for robust feature matching.
• Integrated a layer and block strategy with UAV prior data to enhance accuracy and speed.
• Utilized inter-frame matching information to address registration unevenness.
• Introduced a novel approach of updating map features with UAV image features for improved robustness.

Main Results:

• Achieved stable and accurate registration of UAV aerial images onto maps.
• Demonstrated adaptability to camera head and environmental changes.
• Reached a frame rate of 12 frames per second, suitable for real-time applications.

Conclusions:

• The proposed method provides a feasible solution for real-time UAV aerial image and map registration.
• The approach enhances robustness and applicability in dynamic environments.
• Establishes a foundation for precise geo-positioning of UAV aerial image targets.