Progress and Prospect of LiDAR Point Clouds to 3D Tree Models

3D geometric tree models are of great interest to many applications, such as digital city and digital forestry, among others. Of late, light detection and ranging (LiDAR) technique has been extensively used to capture the geometric shapes of the trees from the outdoor scenes. Despite two decades of research, tree modeling algorithms and the created tree models are still far from being satisfactory. In this paper, we review most of the mainstream tree modeling algorithms by using ubiquitous point clouds. These tree modeling algorithms can be roughly classified into five categories, including clustering-based method, graph-based method, a priori assumption-based method, Laplace's method, and lightweight expression-based method. In each category, we analyze the strengths and challenges of the tree modeling algorithms. Afterwards, some possible tree modeling methods and strategies are given to overcome the potential limitations in terms of detailed skeleton representation, robustness and scalability, level of details (LoDs) representation, and tree modeling evaluation. We finally propose a few suggestions for future research topics in tree modeling.