A Registration Algorithm Of Plant Point Clouds Based On Geometrical Feature Constraint Of Neighborhood

Virtual plant model plays an important role in agriculture and accurate plant 3D point cloud is important to rebuild a plant model. Three-dimensional point cloud registration is important in reverse engineering. The complete 3D data model of physical surface cannot be captured from a viewpoint since the irregular size and shape about the physical model. In order to get a complete point cloud data of surface of model, it is necessary to get point cloud data from different viewpoint, and register these point cloud data.The methods of manual registration labeling and making points in model and of calibrating the relative movement are adopted by current commercial products to register point cloud, but the registration algorithm cannot meet the need of the project because of the limited application scenarios. In order to meet the needs of this project, however,this paper aims to provide automate registration program for the 3D Scanner researched and developed by laboratory based on the features and implement environment. In this paper, a registration method for large-scale 3D point clouds is proposed by the in-depth study of the three-dimensional point cloud registration technology, which is based on neighborhood constraints of geometrical features. The method consists of initial and exact registration steps. The main contents and innovations are as follows:(1) In the process of initial registration, we defined a new function that measures feature similarity by calculating the distance function, and the method based on feature similarity of point cloud data was proposed to implement the initial registration. Firstly,for the laser point cloud is scattered and lacks the topological information, the extraction of the geometrical feature of point cloud is difficult. So, this paper analyzed the impact of the size of the neighborhood on the surface fitting selecting the k-neighborhood based on real environment to fit its support surface, so that the geometric feature is calculated further. Secondly, For the problem of time consuming in the process of selecting the matching points caused by the large-scale point cloud data, this paper constructed the relationship table of point, neighborhood relations and geometric feature between sourcepoint cloud and target point cloud, and weighted and sampled them to compress the search space and reduce the time-consuming to find matching points. Finally, for the problem that how to choose the high accurate matching points, this paper selects the matching feature point pairs by defining five dimensional feature space and similarity function of feature excluding a large number of false matching points and calculating the coordinate transformation parameters.(2)The iterative closest point algorithm was improved to increase the accuracy in the process of exact registration. After initial registration, the point cloud data was converted to the unified coordinate systems. In order to further increase the accuracy of point cloud registration, the two new angle features was added to improve the result of initial registration achieving the exact registration and acquiring the complete point cloud model of object.Compared with the traditional feature-based and iterative closest point algorithms,the method proposed by this paper significantly reduced the registration time by 11.9%and has only 1% of the registration error of the traditional feature-based algorithm. The proposed algorithm can be used to create efficient 3D models for virtual plant reconstruction and computer-aided design, and the registration results can provide a reference for virtual plant reconstruction and growth.