A marker-free method for registering multi-scan terrestrial laser scanning data in forest environments
A marker-free method for registering multi-scan terrestrial laser scanning data in forest environments
复制标题
森林环境中多扫描地面激光扫描数据的无标记配准方法
DOI：
10.1016/j.isprsjprs.2020.06.002
复制
发表时间：
2020-08-01
影响因子：
12.7
通讯作者：
Guo, Qinghua
中科院分区：
文献类型：
--
作者：
Guan, Hongcan;Su, Yanjun;Guo, Qinghua
地面激光扫描（TLS）已被公认为一种准确的手段，非破坏性地获得三维（3D）森林结构属性。这些属性包括但不限于树高、胸径和叶面积密度。因此，TLS已成为森林调查实践和森林生态系统研究中日益重要的技术。在对林分冠层三维结构进行综合表征时，往往需要在不同位置采集多个TLS扫描数据，其中多扫描配准是一个关键的前提条件。目前，森林中的多扫描TLS注册主要基于非常耗时和繁琐的过程，即在现场设置手工制作的注册目标，并从收集的数据中手动识别扫描之间的共同目标。在这项研究中，提出了一种新的无标记的方法，自动注册多扫描TLS数据。该方法的主要原理是从单次TLS扫描的原始点云中识别阴影区域，并将其用作注册多次扫描TLS数据的关键特征。该方法是用17对TLS扫描在中国各地的6个样地收集不同的植被特征（如，植被类型、高度和林下植被复杂性）。我们的研究结果表明，该方法成功地注册了所有17对TLS扫描，具有与手动注册方法相同的准确性。此外，该方法省去了现场设置配准目标、人工从TLS数据中识别配准目标以及处理原始TLS数据提取树木个体属性的过程，具有高效性和鲁棒性的优点。预计该算法可以节省在森林中收集TLS数据的时间和成本，从而提高TLS林业应用的效率。
Terrestrial laser scanning (TLS) has been recognized as an accurate means for non-destructively deriving three-dimensional (3D) forest structural attributes. These attributes include but are not limited to tree height, diameter at breast height, and leaf area density. As such, TLS has become an increasingly important technique in forest inventory practices and forest ecosystem studies. Multiple TLS scans collected at different locations are often involved for a comprehensive characterization of 3D canopy structure of a forest stand. Among which, multi-scan registration is a critical prerequisite. Currently, multi-scan TLS registration in forests is mainly based on a very time-consuming and tedious process of setting up hand-crafted registration targets in the field and manually identifying the common targets between scans from the collected data. In this study, a novel marker-free method that automatically registers multi-scan TLS data is presented. The main principle underlying our method is to identify shaded areas from the raw point cloud of a single TLS scan and to use them as the key features to register multi-scan TLS data. The proposed method is tested with 17 pairs of TLS scans collected in six plots across China with various vegetation characteristics (e.g., vegetation type, height, and understory complexity). Our results showed that the proposed method successfully registered all 17 pairs of TLS scans with equivalent accuracy to the manual registration approach. Moreover, the proposed method eliminates the process of setting up registration targets in the field, manually identifying registration targets from TLS data, and processing raw TLS data to extract individual tree attributes, which brings it the advantages of high efficiency and robustness. It is anticipated that the proposed algorithms can save time and cost of collecting TLS data in forests, and therefore improves the efficiency of TLS forestry applications.