Fast and Accurate 3D Reconstruction of Plants Using MVSNet and Multi-View Images

Accurate 3D reconstruction of morphological structures, growth processes is a prerequisite for image-based measurement of biological organisms. It is a particular challenge for digital plant and crop research. In this paper, multi-view images of structurally very different plants including Arabidopsis, Wheat and Physalis, were taken with a consumer-grade digital camera with a zoom lens and a turntable. Camera parameters were estimated using the SfM method (COLMAP), and then 3D point clouds were reconstructed using MVSNet and the camera parameters. The results show that the proposed method is able to quickly produce denser and complete 3D point cloud of plants. Compared with the existing methods, our method is an end-to-end framework and is more automatic and more promising for dense 3D reconstruction of plants, especially for plant phenotyping.