A Scale-Invariant Looming Detector for UAV Return Missions in Power Line Scenarios.

Area of Science:

• Robotics and Automation
• Computer Vision
• Aerospace Engineering

Background:

• Unmanned aerial vehicles (UAVs) are crucial for power grid maintenance but face collision risks with power lines.
• Conventional visual systems and existing deep learning methods struggle with real-time detection of thin power lines in dynamic scenarios.

Purpose of the Study:

• To develop a novel real-time collision avoidance system for UAVs operating near power lines.
• To address the limitations of current methods in detecting intricate power lines and dynamic threats.

Main Methods:

• Proposed a Scale-Invariant Looming Detector (SILD) inspired by biological motion detection (Lobula Giant Movement Detector).
• Implemented frame preprocessing, attention masks for motion enhancement, and simulated biological arousal for threat recognition.
• Compared SILD against Hough transform and D-LinkNet for static power line detection.

Main Results:

• SILD demonstrated a balance between detection accuracy and real-time processing efficiency.
• The system effectively detected looming threats and predicted impending collisions during high-speed flight.
• Achieved consistent sensitivity to looming objects across different scales, overcoming motion vision limitations.

Conclusions:

• SILD is well-suited for UAV-based power line detection, offering high precision and low-latency performance.
• The model was successfully deployed on a UAV-embedded board for real-world collision avoidance testing.
• This approach offers a novel solution for UAV obstacle avoidance in power line environments.