Studies On Automatic 3D Reconstruction Techniques Of Trees Based On Terrestrial Lidar Point Clouds

Trees are one of the most common kind of vegetation in nature, they have a wide catalog and posture, it fully reflects the magical work of nature. With the advancement and development of science and technology, people want and need to reproduce the real scenes of nature in the computer virtual space so that they can be used in scientific research, entertainment and in the life. Because trees exist widely in the nature, so they have become an important part of this study. At the same time, because of their complicated posture, the researches on them are very difficult. With the development of 3D laser scanning technology, precise modeling of branches of trees based on the geometric structure is becoming possible. In this paper, we focus on the automatic reconstruction of trees model based on point cloud data, we have done the following works in the research:1) We design and implement an automatic registration method which completely gets rid of target and manual intervention when we’ve acquired several stations of data with the ground-based laser radar, thus greatly reduces the workload of data collection in the field and simplifies the complexity of data registration.2) Through the research on clustering algorithm, we summed up the point cloud clustering method of local optimization for trees. And by finding the appropriate data structures, we solved the problem of low efficiency in point cloud storage and query. In this paper we use a Kd tree structure to organize and manage our point cloud data, and use the K- mean clustering method to realize the point cloud data segmentation.3) Through a lot of experiments and summaries of different methods, we summarize the skeleton points extraction and skeleton line connection method. This is the core content of our research, because the skeleton structure is the most simple structure which can directly reflect the trend and characteristics of a tree, all our previous work is to get the skeleton structure, and our reconstruction of surface structure is completed also based on skeleton structure. Skeleton structure contains four key parameters: skeleton point position, radius, directions and the connection of skeleton points. This paper gives the detailed solution for four key parameters and experimental results, and compares and analyses our various implemented method for some key part.4) Through the research of skeleton points and skeleton line parameters, this paper realized the resampling process of a point cloud data. And according to the characteristics of the resampled data, we explored and realized the connection relations between points, and finally generated the tree’s surface structure. This structure is a triangle mesh formed by the regularly connected point cloud data.By analyzing the experimental results and data, our method can obtain the characteristics of trees’ structure realistically. The reconstruction results eventually meet all of our expectations.