基于线扫描数据的复杂环境下的河岸线提取

三维点云边缘提取算法已被广泛用于海岸线的提取中。相较于海岸线，河岸线存在高程变化大、易被树木遮挡等问题，将海岸线提取算法直接应用于河岸线提取存在较多的局限性。本文提出了一种基于线扫描激光点云的复杂环境下河岸线精确提取的方法，实现了河流岸线高程不一致、河岸被树冠遮挡情况下河岸边缘点的精准提取。设计了一种自适应高程阈值动态调整方法，通过对高程进行实时统计，使得高程阈值能够自适应地变化并克服树木遮挡造成的岸线不连续的问题，以此获取更加真实的岸线信息。同时，将法向渐变作为约束条件，能在实现对提取的边缘点云滤波的同时构造出完整、平滑的河岸矢量线。最后，利用真实无人机载激光雷达数据验证了本文所提方法的有效性和精度。本文方法获取的岸线与真实岸线的均方根误差分别为0.2018 m和0.2675 m，本文方法获取的河岸线具有较高的完整性。本研究为复杂情况下的河岸边缘提取提供了一种高效精确的方法。

Abstract

Objective

In recent years, with the rapid development of LiDAR, algorithms for point cloud edge extraction have been widely used in coastline extraction. Compared to coastline extraction, riparian line extraction has more significance; however, it often presents more challenges, especially when confronted with the task of extracting riparian lines in environments characterized by substantial elevation variations and dense tree cover. To address these problems, we propose an accurate extraction process for riparian lines in complex environments using a line-scanning laser point cloud. This approach enables precise riparian line extraction even in scenarios where the riverbank elevation varies significantly and the riverbank is obscured by tree canopies.

Methods

First, the original river point cloud data were preprocessed, and the number of points within a certain range was counted using the center point neighborhood and elevation information. The center points were then categorized into noise and riverbank points, and an initial representation of the riverbank was established using the breakpoint analysis method (Fig.2). Second, an adaptive threshold frame was generated. With the rough outline as the reference and the rough outline point as the center point of the threshold frame, elevation statistics were performed on the object elevation within the threshold frame. Concurrently, ground objects, such as tall trees, were selectively removed to refine the threshold frame (Fig.11). Finally, considering the smoothness characteristics of the riparian edge in real conditions, “defective” points were removed by the normal gradient constraint between adjacent edge points. This ensured the smoothness of the edge, and the remaining edge points were subsequently connected to form vector riparian lines. The detailed process for precisely extracting a riparian line in a complex environment is shown in Fig. 1.

Results and Discussions

Experiments were conducted on islets in the middle of a river and shorelines obscured by trees using the method described in this study and the mainstream contour tracing method, respectively. The shorelines extracted using the two methods were compared with real shorelines. The root mean square error (RMSE) of the calculated distances was used to qualitatively analyze the performance of the two methods. As shown in Table 1, the proposed method exhibit an approximately 24.6% lower RMSE compared to the contour tracing method, along with 30.2% more error-free matches. Moreover, it is evident from the statistical histogram of the error distribution that the errors of the proposed method are mainly concentrated within 0.3 m, with only a few points exceeding 1 m (Fig.16). Although the error of the contour line tracking method also achieved excellent results, it is significantly larger than that of the proposed method, with certain points exhibiting an error of approximately 1.5 m. For island and reef terrain, both methods demonstrate excellent capability in extracting the bank edge. However, when dealing with coastlines concealed by trees, the edge extracted by the proposed method is more closely aligned with the actual terrain (Fig.17). In summary, the proposed method outperforms the traditional contour tracing method in terms of extraction quality and error control, making it particularly well-suited for processing data in complex terrains.

Conclusions

Considering the complex environment of riverbanks, this study proposes an accurate extraction method for riparian lines using line-scanning laser point clouds in complex environments. The advantages of the method are as follows: (1) it utilizes scan line characteristics to identify breakpoints; (2) an adaptive threshold is employed to eliminate non-riparian point cloud data like shoreline canopy; (3) riparian point cloud data are further transformed into smooth, closed vector data using the normal gradient constraint method. The experimental results show that this method can effectively extract riparian lines in a complex environment. This proposed method holds promise in supporting critical tasks such as river surveys, river track detection, and riverbank soil and water loss monitoring.

1　引言

河岸线是指水陆交界的分界线。河岸线是动态的分界线，在丰水期、平水期、枯水期，水岸线的形状是不同的。关于河岸线不同时期状态的相关研究对于当地的生态环境保护具有重要意义。早期针对河岸线提取的方法多用于遥感领域，但激光点云技术凭借其精度高、速度快、可同时获取大量的点等特性，迅速成为该领域的重点研究对象^［1］。激光点云数据量庞大，对数据处理技术提出了更高要求，进而推动了针对激光点云的去噪、分割、边缘提取等算法的开发与研究^［2-5］。搭载在飞行器上的激光雷达可以在极高的视角下实现对地面信息的高精度、大范围获取，在地物边缘提取任务中得到了极为广泛的应用^［6］。苗松等^［7］针对扫描过程中存在的建筑物遮挡问题，提出了数据补充的方法，并采用该方法提取了被遮挡部分的边缘信息。陈华伟等^［8］在特征线提取过程中引入图像处理的思想，提出了鲁棒性极强的基于曲率突变分析的特征线提取方法。高佳月等^［9］提出一种基于局部边缘特征描述子的边缘检测算法，该算法在很大程度上解决了当前点云边缘检测算法由邻域半径设置造成的边缘提取结果不稳定的问题。王明军等^［10］提出了一种基于自适应局部邻域特征点提取和匹配的点云配准方法，实现了特征点的高效提取。Wang等^［11］搭建了边缘覆盖卷积神经网络（SC-CNN），该网络在获取点云数据边缘特征的同时实现了语义层面的分割。李维刚等^［12］结合道路结构特征和点云回波强度信息提出了一种道路标线提取方法。

由于水体与岸上地物对电磁波的反射特性不同，激光点云在海岸线的提取中得到了广泛应用^［13］。董保根等^［14］先利用离散LiDAR点云数据构建约束三角网，然后结合样条函数内插实现了平滑海岸线的提取。于彩霞等^［15］提出了基于点云栅格化的海岸线提取方法，该方法简洁高效地实现了海岸线的提取。后来，他们又在LiDAR数据的基础上^［16］，将潮水面作为阈值，实现了对海岸边点云数据的水陆分割，得到了可靠性更高的海岸线结果。刘森波等^［17］利用等值线追踪法实现了海岛岸线的高精度、高效率、低成本提取，为海岸线提取方法的研究提供了新的思路。点云数据用于海岸线提取已经获得长足发展，但在进行河流岸线的提取时往往会存在诸多问题：① 河流所处的地形较为复杂，河岸的高程往往存在较大变化，从而使得经典的等值线追踪法很难得到优异的结果；② 海岸周围往往地物较为简单，但河岸周围茂密的树木会遮挡激光扫描仪对河岸线的扫描；③ 河岸线曲折蜿蜒，从所提取的点云中还原真实的河岸走向存在困难，获取的河岸线点云数据在后续应用中也较为不便。笔者通过分析复杂环境下河岸点云的特征，提出了一种基于线扫描激光点云的复杂环境下河岸线提取框架。然后，从点云数据预处理、边缘提取等方面入手，提出了自适应动态阈值方法，并采用该方法将树冠与河岸区分开来，避免了其干扰河岸线的提取。之后结合法线方向渐变约束方法，实现了所提取的精确河岸点云的矢量化，为后续研究和应用提供了支持。

2　河岸线提取方法

河道边缘环境复杂，存在灌木、树林、水草等多种地物，这些地物会对河岸线的精准提取带来一定干扰。尤其是高大的树木，其树冠延伸到水面，遮挡了三维激光扫描对河岸数据的获取。因此，亟须设计出一种合理完善的解决方法，用于实现树木遮挡情况下河流岸线的准确提取。本文提出了如图1所示的基于线扫描的复杂环境下河岸线提取框架，其总体上分为三部分。首先，对原始河道点云数据进行预处理，利用中心点邻域信息和高程信息统计一定范围内点的个数，将中心点划分为噪声点和河岸点，利用断点分析法获取河岸的粗略轮廓；然后，生成自适应阈值框，以粗轮廓为引线，以粗轮廓点为阈值框中心点，对阈值框内的地物进行高程统计，有选择性地去除高大的树木等地物；最后，考虑到真实情况下河岸边缘的平滑性特点，利用相邻边缘点之间的法向渐变约束去除“瑕疵”点，以保证河岸边缘平滑，然后将边缘点连接为矢量河岸线。