In order to achieve the point cloud acquisition of cucumber plants in greenhouse tall crops, a dense map building algorithm was proposed. The algorithm was based on the ORB-SLAM3 algorithm architecture. Firstly, by improving the extraction process of feature points, the quadtree extraction method was used to make the distribution of feature points more uniform and improve the quality of key points. Secondly, it added dense map building thread, octree map thread and raster map thread. The dense mapping thread usually recovered single-frame point clouds and combined statistical filtering and voxel filtering, and then transferred the cucumber point clouds from the camera coordinate system to the world coordinate system for alignment and fusion according to the camera poses on both sides of the cucumber plants. Compared with the traditional rotary multi-view alignment method, it solved the problem of missing alignment information of the point clouds on both sides of the ridge, and successfully achieved the automatic alignment and fusion of the point clouds on both sides of the ridge, and finally obtained a high-accuracy greenhouse point cloud. The algorithm solved the problem of missing information in the point cloud on both sides of the ridge, and successfully achieved the automatic alignment of cucumber point clouds on both sides of the ridge. In order to verify the practicality, the TUM dataset and the real scene were tested, and the results showed that the enhanced ORB-SLAM3 algorithm was more accurate in running trajectory, and its absolute error was reduced by 21.4% on average. The research achieved three-dimensional point cloud acquisition of tall fescue crops and provided basic data for the subsequent analysis of phenotypic data.