Efficient Autonomous Drone Landing: A Computer Vision Approach Optimized for Low-Power Platforms

The rapid advancements in Artificial Intelligence (AI) have significantly transformed the fields of robotics and unmanned aerial vehicles(UAV). Modern drones are now equipped with sophisticated path planning and navigation systems. However, achieving fully autonomous flight remains a challenge due to the need for human intervention at various stages, particularly during landing. This human dependency limits the potential of drone application and usage in industries. An autonomous drone system must integrate multiple components, including security checks, path planning, navigation, detection, validation of landing locations, identification of landing areas, and return to its launch pad after completion of the task. Among these, autonomous landing is the most critical yet unresolved challenge. Detection of suitable landing sites requires substantial computational power and memory, which is limited in drones. This paper proposes a novel approach for detecting autonomous landing sites using computer vision-based techniques. By leveraging these techniques, our approach optimizes it for limited computational resources contrary to the neural networks or deep learning models. The proposed architecture is physically validated through deployment on the Jetson Nano platform, demonstrating its feasibility and effectiveness.