Abstract: Abstract: A virtual plant growth model is very important to agriculture and a crop growth mechanism model. The precise three-dimensional morphological structure of the plant model can be used to study the spatial structure associated with the nature of plants. The acquisition of a plant 3D point cloud is the first step to establish a plant 3D model. In this paper, the registration method and visualization method of the plant point cloud were mainly researched. At first, the research status of virtual plants and plant three-dimensional visualization was introduced, and the feasibility and necessity of the three-dimensional visualization of plant leaves was discussed. Then, a point normal calculation method of registration sphere was studied. Ensemble technology was used to improve the accuracy of the point normal value. By this technology, the normal value of the point cloud was substituted by the normal average value. This was done so that more accurate sphere centers could be calculated and we could get more accurate rotation shaft which was necessarily to register 3D points acquired from different sides. The outlier points and noise points of the original 3D point cloud can be removed at the same time. The point normal value is closely related to the number of point neighbors which are involved in the point normal calculation. So the effect of the number of point neighbors on calculating point normal was discussed and compared. The Moving Lease Squares (MLS) was used to fit the surface of the registration sphere, and Gaussian curvature was calculated to recheck the accuracy of point normals. Two aluminum alloy elements were designed to evaluate the accuracy of the 3D point cloud collection equipment. Moreover, to improve the registration effect of plant leaves, the iterative closest point (ICP) algorithm was used. At last, a point cloud processing system was developed based on Microsoft Visual Studio C++ 2010. The open source development toolkit: Visualization Toolkit was used to realize the 3D point cloud visualization effect and registration algorithms.