Monocular Visual-SLAM for Occluded Tomato Detection in Greenhouses via Hierarchical Localization and GLOMAP
Original title:Visual-SLAM for the detection of hidden tomatoes in greenhouses by Hierarchical Localization and GLOMAPfor robotized harvesting
Detecting fruits occluded by dense foliage inside greenhouses typically relies on costly LiDAR or stereo sensor setups. This paper introduces a cost-effective monocular Visual-SLAM pipeline running on ROS 2 Humble, integrating the Hierarchical Localization coarse-to-fine framework with the GLOMAP Structure-from-Motion mapper. Tested on real tomato clusters in an experimental greenhouse, the system reconstructs occluded fruits that conventional vision systems frequently miss. Comparisons against manual ground-truth measurements of size, centroid position, and orientation confirm the geometric reliability of low-cost monocular pipelines for agricultural harvesting robots.
Why it's worth reading
Demonstrates a practical, low-cost monocular SLAM approach using GLOMAP and hierarchical localization to reliably reconstruct occluded crops for agricultural harvesting robotics.