A research team led by Professor Kyeong-Hwan Lee at Chonnam National University addressed the limitations of existing mapping techniques by creating a system that merges aerial remote-sensing imagery with ground-based LiDAR-inertial odometry. While drones provide a broad aerial perspective, they struggle to capture details beneath thick foliage. Conversely, ground robots generate high-fidelity 3D point clouds but lose accuracy over long distances as satellite signals weaken within dense rows of trees.
In section Releases
Chonnam National University AI Maps Orchards with Centimeter Precision
Navigating dense orchards remains a persistent hurdle for autonomous agricultural robots, as tree canopies often obscure satellite signals and cause ground-based sensors to drift. A new cross-modal fusion framework developed by South Korean researchers now bridges this gap by aligning low-altitude drone imagery with ground-level LiDAR data.
The team’s solution, detailed in Artificial Intelligence in Agriculture, utilizes a deep learning model to align these data streams. By converting LiDAR data into a structured 2D Bird's Eye View map, the system matches it against aerial imagery using pixel-level structural cues. This process suppresses drift and allows for the construction of detailed, geographic information system-driven models. Field tests in apple orchards confirmed the system achieves centimeter-scale localization accuracy, remaining robust against seasonal changes and structural variations. Designed for deployment on embedded devices, the technology provides a foundation for real-time robotic navigation and health monitoring in commercial farming.
Comments (0)
No comments yet. Be the first!