Verlag der Österreichischen Akademie der Wissenschaften

Abstract:
In viticulture, yield prediction plays an important role, helping winegrowers to predict the start of the next growth stage of vines and to improve vineyard management decision-making. To predict a vineyard’s yield, it is necessary to gather accurate local information about the vine’s phenology and morphology, such as the volume of individual grapes. Traditional collection of these data and yield prediction rely on resource- and time-intensive direct visual and manual in-field work by viticulturists. Thus, only limited sampling in the vineyards is possible, carried out by humans. Automated procedures utilizing sensor-based systems reduce the data acquisition time and enable the collection of high-resolution data from the entire vineyard. Large-scale 3D models of vineyards can be generated from these data and used to analyse, for example, the vineyard’s yield or the vegetative stage of individual vines. We propose a concept for a 3D model that uses close-range photogrammetry. In a laboratory experiment, we tested the acquisition of multi-view image datasets from grapes using close-range photogrammetry and derived physical and morphological parameters from 3D grape models. The results could contribute to the design and implementation of a large-scale in-field experiment.