A Visual Positioning Method of UAV in a Large-Scale Outdoor Environment

Area of Science:

• Robotics and Computer Vision
• Artificial Intelligence

Background:

• Structure-based visual positioning is crucial for Unmanned Aerial Vehicle (UAV) operation.
• Existing methods struggle with illumination and seasonal changes, leading to unreliable localization and slow performance with large databases.

Purpose of the Study:

• To develop a novel hierarchical visual positioning method for UAVs.
• To improve the efficiency and accuracy of visual positioning under challenging environmental conditions.

Main Methods:

• A cognitive map is constructed using brain-inspired mechanisms and landmarks for efficient image retrieval.
• Graph neural networks learn feature point relationships, incorporating semantic confidence to enhance matching accuracy.
• A PnP solver is used for pose calculation, with mismatch elimination by analyzing landmark-based matching results.

Main Results:

• The proposed method significantly reduces image retrieval time by two-thirds.
• Achieved an average positioning error of 10.8 meters in simulation datasets.
• The matching algorithm demonstrated superior performance in specific challenging scenes.

Conclusions:

• The hierarchical visual positioning method offers a robust and efficient solution for UAV navigation.
• The integration of cognitive maps and graph neural networks enhances localization accuracy and speed.
• This approach addresses key limitations of traditional visual positioning systems in dynamic environments.