Development of High-resolution 3D Phenotyping System Using Depth Camera and RGB Camera

Abstract. The importance of phenomics which is a research field covering the phenotype such as physiology, morphology and biochemical characteristics of crops is emerging, following the development of convergence technology for plant analysis. Therefore, analysis of 3D images beyond 2D images is required for accurate analysis of crops. In this study, system configuration and algorithm implementation for 3D reconstruction of red pepper were performed. A turntable was used to acquire multiple images for 3D reconstruction, and a method of rotation of plants by turntable was selected. Multiple images were acquired at equal time intervals via a fixed position camera while the plant rotated once. In order to obtain more accurate 3D reconstruction images, Kinect v2 with depth sensor and a high resolution RGB camera were used. Attained 3D reconstruction images were compared with conventional reconstruction images of the existing program using structure from motion (SfM) algorithm, in terms of accuracy. As a result of the comparison, the algorithm implemented in this study showed better overall performance. In addition, the reconstruction image was compared against a ground truth data, which directly measured features such as the number of leaf, width, and plant height according to the growth stages of the plant. It shows the method has good performance to reconstruct 3D images and is suitable for high throughput phenotyping. Acquired images by the 3D reconstruction method can be applied to various fields such as various image and image segmentation.