Estimation of Walnut Structure Parameters Using Terrestrial Photogrammetry Based on Structure-from-Motion (SfM)

Abstract

Remote sensing techniques are increasingly used for crop monitoring to improve the profitability of plantations. These studies are mainly based on spectral information recorded by satellites or unmanned aerial vehicles. However, the development of Earth Observation Systems capable of retrieving 3D point clouds at an affordable cost enables the possibility of exploring new approaches in agriculture. In this context, more research is required to analyze the capability of 3D data for inventory, management and prediction of inputs (water, fertilizers and pesticides) and outputs (production, biomass) of fruit plantations. To do this, the complete representation of each tree contribute to extract the main geometric parameters. The objective of this work is to obtain regression models to estimate total height (Ht), crown height (Hc), stem diameter (Ds), crown diameter (Dc), stem volume (Vs) and crown volume (Vc) from 45 walnut specimens. For this, 3D models were computed for these trees by applying ground-based Structure from Motion (SfM). A circular photogrammetric survey of each tree was carried out using a standard digital camera and three-dimensional point clouds were retrieved for each tree. From these data, the tree parameters were computed. Linear regression models were obtained to estimate Ht, Hc, Ds, Dc, Vs and Vc, with R2 values between 0.89 and 0.99. The results showed accurate fits between field parameters and those derived from 3D point clouds retrieved from SfM technique, indicating the applicability of this cost-effective method to model walnut trees and to extract their accurate parameters without costly field campaigns.