Reconstruction of 3D Point Cloud of Entire Greenhouse Using RGB-D Camera

Research in computer vision for agriculture has been vigorously pursued. To stabilize yields and increase revenues using smart agricultural technologies, it is necessary to enhance the efficiency and accuracy of crop condition diagnostics. However, existing point cloud construction techniques primarily aim to generate point clouds at the pot level, and applying them to an entire greenhouse involves difficulties in alignment and challenges in outdoor imaging. In this study, we developed scanning and preprocessing methods that mitigate these issues. We also developed a technique to extract the parameters necessary for growth diagnostics by analyzing the 3D point clouds of the digital twin greenhouse constructed. The utility of the developed system was validated in a greenhouse handling bell peppers. The results confirmed that the parameters necessary for the growth diagnostics targeted in this study meet the requirements of growth diagnostics, both spatially and temporally. Future prospects include predicting harvest volumes and visualizing know-how using data obtained from the digital twin greenhouse.