Plant growth parameter estimation from sparse 3D reconstruction based on highly-textured feature points

Crop canopy spatial parameters are indicative of plant phenological growth stage and physiological condition, and their estimation is therefore of great interest for modeling and precision agriculture practices. Rapid increases in computing power have made stereovision models an attractive alternative to common single-image-based 2D methods, by allowing detailed estimation of the plant's growth parameters regardless of imaging conditions. Models that have been proposed thus far are still limited in their application because of sensitivity to outdoor illumination conditions and the inherent difficulty in modeling complex plant shapes using only radiometric information. Assuming that not all of the plant-related pixels are essential for growth estimation, this study proposes a 3D reconstruction model that focuses on selected salient features on the plant surface, which are sufficient for obtaining growth characteristics. In addition, by introducing a hue-invariant model, the proposed algorithm shows robustness to diverse outdoor illumination conditions. The algorithm was tested under greenhouse and field conditions on corn, cotton, sunflower, tomato and black nightshade plants, from young seedlings to fully developed plant growth stages, and accurately estimated height (error similar to 4.5 %) and leaf cover area (error similar to 5 %). Furthermore, a strong correlation (r(2) similar to 0.92) was found between the plant's estimated volume and measured biomass, yielding an accurate biomass estimator in the validation tests (error similar to 4.5 %). This estimation ability remained stable while applying the model on plants with varying densities (overlapping leaves) and imaging setups where the standard 2D based analyses failed, thus showing the 3D modeling contribution to robust growth estimation models.