Simulation of tree point cloud based on the ray-tracing algorithm and three-dimensional tree model

Highlights

• A tree point cloud simulation method is proposed.

• The goal is to reduce the uncertainty and cost of field scan.

• The simulation is based on the ray-tracing method and tree models.

• The simulation method is verified by using six tree models coming from two sources.

• The accuracy is analyzed by using the abstract Hausdorff distance.

A terrestrial laser scanning (TLS) instrument can scan trees to get three-dimensional (3D) point clouds with detailed tree structures. However, some factors, such as surrounding environment, tree structure, the characteristics and scanning parameters of TLS instrument, can affect the scanning result. Simulation of tree point cloud is an efficient way to avoid and analyse the influence of the above factors. A simulation method was proposed to simulate tree point clouds by using the ray-tracing algorithm and 3D tree models. Thirteen tree models were tested with the proposed simulation method. Ten tree models are from the virtual model library, and other three tree models are constructed from real point cloud data. The accuracy of the simulated results was evaluated by using the absolute value of Hausdorff distances (AVOHD). Although the results showed that leaves and more branches will decrease simulation accuracy, the proposed method is effective and accurate on tree point cloud simulation.

Introduction

The ecological benefits of forests have non-negligible influences on human beings (Pan et al., 2013). Tree structure and its changes are important information for forest inventory, such as the diameter at the breast height (DBH) (Antonarakis & Alexander, 2011; Simonse et al., 2003), tree height (Hopkinson et al., 2004; Kenneth, Johan, Håkan, 2014), and leaf area index (LAI) (HIROSE, 2005; Ma et al., 2016) etc.

TLS instruments can scan the object quickly and directly, and meanwhile obtain the 3D information, which is called the point cloud data. Due to including the 3D spatial and surface characteristics of the object, the point cloud data provides a new tool for forestry research (Lim et al., 2003). Therefore, the point clouds of trees have been widely used to estimate some tree parameters, such as the diameter at breast height (DBH) (Antonarakis & Alexander, 2011; Chmielewski et al., 2010), tree height (Kenneth, Johan, Håkan, 2014; Monika & Zheng, 2011), leaf area index (LAI) (Béland et al., 2011; Ma et al., 2016), and above-ground biomass (Tian et al., 2011; Gonzalez de Tanago et al., 2018).

In practice, the acquisition of tree point cloud data is often affected by factors such as the characteristics of instrument, scanning parameters, tree structure and the surrounding environment (Wang et al., 2015). Sometimes, it may be not possible to get close to some trees to obtain the point cloud data (Hilker et al., 2010). And weather change during the scanning will also affect the results, for example, wind blowing on twigs or leaves will also result in noisy data (Yun et al., 2019). The structural complexity of a tropical forest can potentially have large impacts on the acquired TLS data (Gonzalez de Tanago et al., 2018).

And the scanning parameters may also be changed for different applications and result in different amounts of data and different costs of time and computation. For example, the required detail level of point cloud data is different for the DBH estimation and the branch distribution description.

In addition to the field data collection issues mentioned above, multiple trials and repeated scanning are often inevitable in scientific research. However, repeated or re-entrance scan will not only cost a lot in time and human resources, but also introduce uncertainty due to the complex and changing experimental environment.

Point cloud simulation based on tree models is a good solution to this problem. The simulated tree point cloud can be used for preliminary analysis supporting on-site scanning and verification of the measured point cloud. It will be a great help to analyse different scanning parameters and find out more appropriate parameter settings in advance.

Teobaldelli et al. (2008) reconstructed 3D tree models and extracted tree topology and geometric parameters manually based on leafless poplar forest data obtained by using a portable laser scanner. Yan et al. (2009) proposed a tree reconstruction method of point cloud data based on cylinder modelling. Michael et al. (2004) used a series of continuous cylinders to fit the point cloud of the tree branch and realised the geometric modelling of the tree branch with cylinders. Markku et al. (2017) put forward a method based on an automatic quantitative model to identify tree species from the point cloud acquired by TLS accurately and automatically. Hackenberg et al. (2015) studied a simple-tree algorithm by using point cloud data to construct 3D models of trees.

Meanwhile, point cloud simulation has attracted more attention due to the cost reduction of time and labour. Furthermore, compared to in-situ data from TLS instruments, the simulated point cloud data has good repeatability to avoid the impacts of TLS instruments, trees and environmental variations. According to the reference investigation, point cloud simulations can be divided into two categories. One category is simulating some parts of trees. Bu & Wang, 2016 discussed the impacts of different angle resolutions based on artificial data simulated by the inclined tree cylinder model. Forsman et al. (2018) analysed the deviations of the stem diameter estimation under different beam widths by using a laser multi-parameter simulator. Wang et al. (2019) investigated the effects of error parameters, scanning parameters and estimation methods on the accuracy and speed of DBH estimation by using simulated point cloud slices. This type of simulation method focused on simulating the point cloud of the inclined trunk slice, instead of simulating the entire tree. Thus, this kind of simulation method was not comprehensive enough to describe the tree in detail. The other category is simulating the 3D tree model. Wang et al. (2015) presented a ray-tracing algorithm to simulate the interaction between laser and single wood and discussed the factors affecting the simulation process. Yun et al. (2019) introduced a new multi-platform Lidar simulation of the 3D tree model to analyse the impact of occlusion on the Lidar-based leaf area by collecting in-situ measurement data from various canopies. However, the tree models were mainly related to the total leaf area of the tree canopy. In other words, the simulation paid attention to specific tree parameters.

However, neither of the above-mentioned simulation research studies involved a combination of tree models and scanning parameters to help in-depth analysis of parameter influence. Similarly, sampling the tree models simply to get the point cloud cannot reflect the effects of the TLS parameters. So, this paper proposes a tree point cloud simulation method which is based on the ray-tracing algorithm and two different kinds of tree models. One kind of tree model is from SpeedTree (2002), which is a virtual vegetation software created by Interactive Data Visualization, Inc. The other kind of tree model is from field-measured point cloud of real trees (FMPRT).

The paper is organised as follows: Section 2 introduces the proposed simulation method. Section 3 describes the simulated results by using six tree models from different data sources. Section 4 analyses the simulated results and discusses the advantages and deficiencies of the simulation method. Section 5 summarises the characteristics of the method and plans future work.