Problem
Leaf area is one of the most used plant traits, and one of the most tedious to measure: the standard method is to cut the leaves off and scan them flat. A 3D reconstruction of the intact plant can give the same number without destroying it, but only if the leaf surface it produces is clean enough to measure.
Approach
The pipeline reconstructs each plant from multi-view images, then extracts the leaf blades as surfaces from both MVS point clouds and Gaussian surfels. The paper focuses on two steps that decide the accuracy of the final number: separating the blade from its supporting structure, and turning the reconstruction into a measurable surface.
Validation
Estimated leaf areas are compared against destructive measurements of the same plants.
Publication
Blade support and surface extraction in soybean leaf-area measurement using MVS and Gaussian surfels. Under review at Computers and Electronics in Agriculture.