SlimDeblurGAN-Based Motion Deblurring and Marker Detection for Autonomous Drone Landing

Area of Science:

• Robotics and Automation
• Computer Vision
• Artificial Intelligence

Background:

• Autonomous drone landing relies on accurate marker detection.
• Existing methods struggle with non-uniform motion-blurred images.
• Deep learning offers superior detection but requires optimization for real-time applications.

Purpose of the Study:

• To develop a robust deep learning-based marker detection system for autonomous drone landing.
• To address the challenge of non-uniform motion-blurred input images.
• To optimize the system for a balance between processing time and detection accuracy.

Main Methods:

• A two-phase framework combining deblurring and object detection.
• Utilized a slimmed Deblur Generative Adversarial Network (DeblurGAN) for deblurring.
• Employed a You Only Look Once version 2 (YOLOv2) detector for marker identification.
• Introduced SlimDeblurGAN, a channel-pruning framework for model optimization.

Main Results:

• The proposed method demonstrated superior performance and robustness compared to existing techniques.
• Effective handling of non-uniform motion-blurred images was achieved.
• The SlimDeblurGAN model maintained high accuracy while reducing processing time.

Conclusions:

• The developed method provides a significant advancement in autonomous drone landing systems.
• The approach offers a practical solution for real-world applications with challenging visual conditions.
• Optimized deep learning models are crucial for efficient and accurate drone navigation.