Comparing NeRF and LiDAR-Based Plant Reconstruction

Despite Structure from Motion (SfM) photogrammetry has already proved to be a very effective tool for 3D reconstructions in a wide range of operational conditions, it still has some flaws when working in certain critical cases. Among such cases it is possible to mention also the 3D reconstruction of crops and plants, which can be of interest in some developments of precision agriculture. The criticality in this case is usually mostly related to the difficulty in reliably matching conjugate points, mostly because of the high level of similarity between different leaves. The quest for ever improving precision agriculture solutions motivates the search for alternative plant 3D reconstruction techniques. Neural radiance fields (NeRF) has proven during the last few years to be a very effective solution in several conditions challenging for standard SfM photogrammetry. Hence, this work aims at assessing the geometric reliability, in terms of completeness and geometric accuracy, of NeRFbased 3D reconstruction of plants, by providing a comparison with laser scanning-based reconstructions, considered here as reference solutions, i.e. used here for generating the ground truth geometry of the plants. NeRF-based metric reconstructions are obtained taking advantage of a low-cost GNSS receiver attached to the camera.