Answers are grounded in this paper and cite their sources.
This article presents a reproducible hybrid 2D-3D RGB-D method for estimating the growth axis of Phalaenopsis orchid buds and converting it into a robotic grasp pose. Synchronized RGB images and depth maps are used as input. Instance segmentation first isolates individual buds, and the resulting masks are used for 2D skeleton extraction and mask-constrained 3D point cloud reconstruction. After point cloud denoising, background plane removal, and coordinate normalization, branch-wise Principal Component Analysis estimates local geometric direction candidates. These candidates are fused with skeleton-derived branch topology to determine the final growth axis. The estimated axis is then used to construct a local grasp-oriented coordinate frame and generate a 6-DoF grasp pose in the robot base frame through camera-robot calibration. The method was evaluated on 520 RGB-D samples of Phalaenopsis orchid buds and compared with 2D skeleton-only and global 3D PCA-only baselines. The hybrid method achieved a mean angular error of 5.1° ± 3.3° (mean ± standard deviation), compared with 9.3° ± 4.2° for global 3D PCA and 14.2° ± 6.1° for the 2D skeleton method. It processed each sample in 65 ms on average and achieved a 94.0% grasp success rate (47/50) under controlled laboratory conditions. • The method integrates 2D skeleton topology and 3D point cloud geometry in a reproducible RGB-D processing pipeline. • Branch-wise PCA and skeleton-guided fusion improve growth-axis estimation for multi-branch orchid buds. • The estimated growth axis is converted into a robot-referenced 6-DoF grasp pose for perception-guided manipulation.
No takes yet. Share an insight, caveat, or question.
Nguyen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: