Comparison of Structure-from-Motion and Stereo Vision Techniques for Full In-Field 3D Reconstruction and Phenotyping of Plants: An Investigation in Sunflower

Abstract

Abstract. Accurate full 3D reconstruction of plants is a crucial step in phenotyping the canopy geometry of living plants to measure features, such as plant height, plant volume, leaf count, leaf size, and internode distance. In this work, an in-field, full 3D reconstruction system for plant phenotyping is described. The system utilized simultaneous, multi-view, high-resolution color digital imagery for true 3D reconstruction of the crop. The cameras were mounted on an arc-shaped superstructure and organized into 16 stereo pairs in four separate arcs. In this study, a comparison of structure-from-motion (SfM) and stereo vision (using stereo cameras) techniques in terms of full 3D reconstruction and measurement of plant features is introduced. System performance was verified in outdoor, on-farm experiments conducted at multiple time points in the growth cycle of sunflower. Results show, that both SfM and stereovision techniques can yield satisfactory 3D reconstruction models and are suited for high-throughput phenotyping without destroying any leaves or stems of the plant. By taking into account small details in the plant and edge preservation for leaves, the custom stereovision algorithms for this system could outperform the SfM technique and provided superior 3D reconstruction results.