Greenhouse agricultural technology is undergoing rapid advancements, confronting the challenges of digitization and automation. In this study, we introduce a novel system for real-time 3D reconstruction of greenhouse crops based on SLAM. Employing RGBD cameras, the algorithm captures real-time 3D point cloud frames, subsequently extracting features for pose estimation. Only keyframes are retained in memory, constituting a pose-graph to economize computational resources. This architecture undergoes backend loop closure detection and is subjected to global optimization, further fortified by marker constraints. The resultant optimized pose graph, combined with the keyframes, is synthesized into a holistic 3D point cloud model of the greenhouse environment. Our method highly correlates with manual measurements (R²=0.996, RMSE=6.26 mm). This research offers a robust 3D modeling approach for greenhouse crops, benefiting phenotypic analysis and digital agriculture.