Abstract:

For high-value precision agriculture, monitoring plant growth trends, pest and disease control, and automation processes will standardize operations and increase yield while reducing losses as much as possible. In greenhouse scenarios, one method for monitoring plant growth requires localization, modeling, and post-processing of the plants. However, recognizing and extracting the root position of the plant is difficult for a robotic arm. To overcome this challenge, this paper uses a marker-based localization method to provide the root position directly. After acquiring and iteratively registering the point cloud, the main stem of the plant is extracted for future plant organ segmentation and clustering. Nevertheless, extracting the main stem is a complex task, and although there are studies on skeleton extraction for ordinary trees or wheat, there are few solutions for high-wire plant stem extraction. Therefore, an optimized geometric-based stem extraction algorithm (SEA) can extract the stem point cloud with a high success rate under conditions no matter whether the cloud is intact or the main stem region is occluded.

[Github]