Segmentation and Extraction of Maize Phytomers Using 3D Data Acquired by RGB-D Cameras

Based-field Individual-level crop organ extraction is an important prerequisite for high-throughput crop phenotyping. Phytomer is the organ propagation unit that connects the leaf and stem nodes, and it can be used to measure a variety of morphological phenotypic parameters. However, the severe occlusion across neighboring crops in dense field scenes leads to difficulties for crop organ point cloud extraction. To tackle this challenge, we study phytomer extraction methods based on point density features. In this paper, we propose a point cloud segmentation approach for individual-level crop organs based on point density histograms and cubic B-sample curve fitting, and also propose a phytomer construction approach based on point cloud color information and density features. To verify the effectiveness of our approach, we conduct field experiments under open-field conditions with a mobile robot equipped with RGB-D cameras. The experimental results show that the phytomer construction approach based on point density features can accurately extract the organ units reflecting the real crop morphology. Based on the statistical analysis of 15 sets of test datasets, we found that the success rate of phytomer construction is up to 89.2%.