Non-Destructive Measurement of Three-Dimensional Plants Based on Point Cloud

Abstract

In agriculture, information about the spatial distribution of plant growth is valuable for applications. Quantitative study of the characteristics of plants plays an important role in the plants' growth and development research, and non-destructive measurement of the height of plants based on machine vision technology is one of the difficulties. We propose a methodology for three-dimensional reconstruction under growing plants by Kinect v2.0 and explored the measure growth parameters based on three-dimensional (3D) point cloud in this paper. The strategy includes three steps-firstly, preprocessing 3D point cloud data, completing the 3D plant registration through point cloud outlier filtering and surface smooth method; secondly, using the locally convex connected patches method to segment the leaves and stem from the plant model; extracting the feature boundary points from the leaf point cloud, and using the contour extraction algorithm to get the feature boundary lines; finally, calculating the length, width of the leaf by Euclidean distance, and the area of the leaf by surface integral method, measuring the height of plant using the vertical distance technology. The results show that the automatic extraction scheme of plant information is effective and the measurement accuracy meets the need of measurement standard. The established 3D plant model is the key to study the whole plant information, which reduces the inaccuracy of occlusion to the description of leaf shape and conducive to the study of the real plant growth status.

Plant phenotyping relevance

Kinectによる3D点群再構成と葉・茎の分割、葉面積・葉長・葉幅・草丈の自動抽出を開発・評価しており、植物表現型取得が中心である。

Code and data availability

The paper describes Kinect v2.0-based 3D point cloud reconstruction and phenotypic measurement of pepper plants, but contains no data availability statement, no public dataset of plant point clouds or measurements, and no author code repository. The only URLs present (PCL resampling tutorial and FLANN kdtree docs) are,

No evidence-backed public reproduction asset is currently recorded.