Multiple Aerial Targets Re-Identification by 2D- and 3D- Kinematics-Based Matching.

Area of Science:

• Computer Vision
• Robotics
• Aerospace Engineering

Background:

• Traditional aerial target tracking often relies on image-based training, which can be computationally intensive and require extensive datasets.
• Re-identification of multiple targets across different sensors presents challenges in maintaining track continuity and accuracy.
• Existing methods may struggle with varying target appearances and complex motion patterns.

Purpose of the Study:

• To develop and evaluate novel kinematics-based techniques for matching and re-identifying multiple aerial targets from multiple electro-optical devices.
• To offer an alternative to traditional image-based training methods by utilizing simpler graph matching algorithms.
• To assess the effectiveness of 2D and 3D kinematics-based approaches in real-world flight test scenarios.

Main Methods:

• A 2-dimensional (2D) method using kinematic and geometric projections of targets onto images, with matching via correlation-based techniques.
• A 3-dimensional (3D) method estimating focal axis motion from target size changes, constructing pose-independent direction vectors for matching in a global coordinate system.
• Both methods facilitate matching and re-identification across frames without explicit overlapping field-of-view requirements, provided targets are detected by multiple cameras.

Main Results:

• The proposed 2D method is effective for targets with distinct observed motion.
• The proposed 3D method is suitable for near to medium-range targets where changes in detection size are observable.
• Preliminary flight tests with 2-3 drones demonstrated the effectiveness of both techniques, achieving a Multiple Object Tracking Accuracy (MOTA) score exceeding 80%.

Conclusions:

• Kinematics-based matching and re-identification offer a viable alternative to image-based training for aerial target tracking.
• The developed 2D and 3D methods provide robust solutions for multi-camera aerial surveillance systems.
• The high MOTA score validates the practical applicability and efficiency of these novel tracking techniques.