YOLOv5 Drone Detection Using Multimodal Data Registered by the Vicon System.

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• Accurate 3D drone tracking is crucial for applications like surveillance and autonomous navigation.
• Existing detection methods often struggle with precision in complex, multi-camera environments.

Purpose of the Study:

• To develop and evaluate a precise drone detection method for the preliminary stage of 3D drone tracking challenges.
• To introduce a novel evaluation metric for assessing 3D drone detection performance.

Main Methods:

• Training and testing the YOLOv5 deep learning network on real, multimodal data, including synchronized video and motion capture data.
• Utilizing an asymmetric cross with markers for precise 3D position and orientation determination.
• Incorporating synthetic data from the AirSim simulation platform for robust training and testing.

Main Results:

• Demonstrated the effectiveness of YOLOv5 for drone detection in synchronized multi-camera systems.
• Proposed and validated a new metric (average distance between centroids) for evaluating 3D detection accuracy.
• Achieved promising results in drone localization using marker-based cross detection.

Conclusions:

• The YOLOv5 network shows significant potential for precise drone detection in 3D tracking scenarios.
• The proposed evaluation metric offers a more adequate assessment of detection performance in 3D.
• Combining real and simulated data enhances the robustness of drone detection models.