High-Throughput Plant Phenotyping: A Review of 3-D Point Cloud Segmentation and Analysis Techniques

This passage discusses the importance of plant phenotyping and the need for rapid and nondestructive techniques for high-throughput phenotyping. It introduces the concept of plant digitizing, which involves accurate measurements of geometric features in plant specimens. The passage also discusses different methods used for plant digitizing, including contact and contactless approaches such as LiDAR sensing systems, structured light, time-of-flight sensors, and multiview stereo (MVS) reconstruction. It highlights the challenges in segmenting and analyzing 3-D point clouds for plant phenotyping, particularly for dense crops. The passage mentions various methods proposed for leaf segmentation and analysis, but notes that many current methods are not suitable for species with dense canopies and have limitations in terms of automation, accuracy, and real-time performance. The passage then introduces a new approach that involves acquiring 3-D point clouds through MVS, removing stems, and using a region growing algorithm based on multiple features to segment individual leaves in the canopy. Finally, it mentions the extraction of phenotypic traits such as leaf area, length, width, and leaf inclination angle using facet oversegmentation and the minimum.