Drone Detection and Pose Estimation Using Relational Graph Networks.

Area of Science:

• Robotics
• Computer Vision
• Aerospace Engineering

Background:

• The increasing use of Unmanned Aerial Vehicles (UAVs) necessitates advanced drone detection and pose estimation for applications like cooperative flight and low-altitude security.
• Current methods often rely on object detection for position but lack crucial attitude information, hindering adaptive control and cooperative performance.
• Existing 6D pose estimation techniques require extensive pose annotation or 3D models, limiting their applicability to non-cooperative targets.

Purpose of the Study:

• To develop a novel 6D pose estimation algorithm for quadrotors that overcomes the limitations of current methods.
• To enable accurate estimation of drone attitude information for enhanced understanding of their state and intentions.
• To reduce the reliance on precise pose annotations or 3D models, making the algorithm more practical for real-world scenarios.

Main Methods:

• A novel algorithm for quadrotor 6D pose estimation was proposed, utilizing keypoint detection with minimal annotation requirements.
• A relational graph network was employed to process keypoint information and infer pose.
• The Perspective-N-Point (PnP) algorithm was integrated for pose calculation.

Main Results:

• The proposed algorithm achieved state-of-the-art performance in both simulated and real-world scenarios.
• The relational graph network demonstrated inference capabilities for keypoints on the quadrotor's motors.
• Significant improvements in both accuracy and speed were observed compared to existing state-of-the-art keypoint detection algorithms.

Conclusions:

• The developed keypoint-based 6D pose estimation method offers a more efficient and adaptable solution for drone monitoring.
• The approach effectively estimates critical attitude information, crucial for advanced UAV operations.
• This method presents a significant advancement for drone detection and pose estimation in security and cooperative flight contexts.