Metrological Assessment of RGB-D Cameras for 3D Orchard Reconstruction

The deployment of photonic sensors in fruit cultivation significantly enhances crop monitoring and management. Despite their potential, high costs and computational challenges limit their widespread use. RGB-D cameras offer a low-cost alternative, generating real-time point clouds, though with reduced resolution and accuracy. This study introduces a customized methodology to fuse multiple point clouds of an orchard into a comprehensive representation, surpassing the limitations of standard algorithms for this type of data. Metrological validation in an experimental apple orchard was conducted to assess the accuracy and reliability of the generated 3D reconstructions of the plants, comparing a high-performing laser scanner produced by Viametris and Kinect Azure. Results show that Kinect Azure struggles in reconstructing objects smaller than 45 mm, however its performances are comparable with the Viametris system, both in the case of a full double-sided representation of the orchard (fusing the front and back of the orchard in a single point cloud) and of a single view representation (front only).