New offer! Get 30% off one whole year of Unlimited learning. Subscribe for just £249.99 £174.99. New subscribers only. T&Cs apply

# Analysing 3D data

A quick look at analysing 3D plant data

A quick look at how 3D plant model data can be analysed, including the use of skeletonisation and clustering.

## 3D models

We’ve seen how 3D models of plants can be created from multiple views and stored as volumes, point clouds and/or surfaces. Though whole-plant models are useful in themselves, allowing direct computation of global traits like size, volume, etc, many applications require specific plant components to be identified. 3D models provide a rich dataset with which to do this, but it obviously requires further processing. The literature on further analysis of 3D plant models is growing rapidly [1].

## Skeletonisation

Volumetric models are effectively 3D images, and all of the techniques presented in the course for processing and analysis of volumetric images, and more, can be applied to them. One of the most common additional techniques is skeletonization. This is particularly well-suited to elongated structures like plant stems, and aims to identify a single voxel wide path through the centre of the object. The figure below shows a 2D example, for ease of display, but the process transfers straightforwardly to 3D. Here an image of a plant shoot (a) has been converted to grey level (b) thresholded to create a binary (c) image – analogous to a volumetric model – and a distance transform applied (d). Distance transforms compute the shortest distance from each pixel (or voxel) inside a region to the boundary of that region. The effect is to highlight pixels (or voxels) lying at the centre of the viewed object. Identifying strings of highlighted pixels produces a skeleton capturing the core structure of the object (e).

Skeletonisation: a 2D example

## Clustering

The first step in the analysis of point cloud data is often segmentation by clustering. Figure 2 shows some examples of applying a standard clustering algorithm to point cloud data obtained by multi-view stereo [2]. Each cluster is displayed a different colour.
Clustering point cloud data to identify plant components.

[1] Vandenberghe, B., Depuydt, S., & Van Messem, A. (2018, December 10). How to make sense of 3D representations for plant phenotyping: a compendium of processing and analysis techniques. https://doi.org/10.31219/osf.io/r84mk

[2] Pound, M.P., French, A.P., Fozard, J.A. et al. A patch-based approach to 3D plant shoot phenotyping. Machine Vision and Applications 27, 767–779 (2016). https://doi.org/10.1007/s00138-016-0756-8.