3D Shape Reconstruction of Plant Roots in a Cylindrical Tank From Multiview Images

Takeshi Masuda; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019, pp. 0-0

We propose a method for reconstructing a 3D shape of live plant roots submerged in a transparent cylindrical hydroponic tank from multiple-view images for root phenotyping. The proposed method does not assume special devices and careful setups, but the geometry and material of the tank are assumed known. First, we estimate the intrinsic and extrinsic camera parameters by the SfM algorithm, and the scale and axis of the tank are estimated by chamfer matching. Second, we apply the ray tracing considering the refraction for each view, the input images are mapped to the voxels, and then multiview voxels at the same location are robustly merged to reconstruct the 3D shape. Finally, the root feature extracted from the voxel is binarized and thinned by applying the 3D Canny operator. The proposed method was applied to real a dataset, and the reconstruction results are presented.