A Fully Automated Three-Stage Procedure for Spatio-Temporal Leaf Segmentation with Regard to the B-Spline-Based Phenotyping of Cucumber Plants

Abstract

Plant phenotyping deals with the metrological acquisition of plants in order to investigate the impact of environmental factors and a plant’s genotype on its appearance. Phenotyping methods that are used as standard in crop science are often invasive or even destructive. Due to the increase of automation within geodetic measurement systems and with the development of quasi-continuous measurement techniques, geodetic techniques are perfectly suitable for performing automated and non-invasive phenotyping and, hence, are an alternative to standard phenotyping methods. In this contribution, sequentially acquired point clouds of cucumber plants are used to determine the plants’ phenotypes in terms of their leaf areas. The focus of this contribution is on the spatio-temporal segmentation of the acquired point clouds, which automatically groups and tracks those sub point clouds that describe the same leaf. The application on example data sets reveals a successful segmentation of 93% of the leafs. Afterwards, the segmented leaves are approximated by means of B-spline surfaces, which provide the basis for the subsequent determination of the leaf areas. In order to validate the results, the determined leaf areas are compared to results obtained by means of standard methods used in crop science. The investigations reveal consistency of the results with maximal deviations in the determined leaf areas of up to 5%.