Automated Extraction of Road Information from UAV-based Data

When it comes to monitoring the condition of roads, UAV technology can overcome many of the downsides associated with traditional methods, which can be time-consuming, labour-intensive and sometimes subjective. This article explores the opportunities for automated extraction of UAV-based data information about road construction, inventory and road environments.

道路是重要的城市特征之一。他们有效，快速，舒适，安全地相互连接。因此，需要监视他们的当前条件，以确保它们符合标准。但是，监测道路状况的传统方法是耗时，劳动密集型，有时是主观的。监测道路状况的一种相对较新的方法是无人机（无人机或“无人机”）技术。无人机是多个领域增长最快的技术之一，例如精密农业和农业，森林，生态和结构健康监测以及地质，地形和考古图。

UAV-based data is collected as the UAV flies over the study area and captures multiple images. Two different types of flight plans, namely manual and autopilot flight plans, are available in the UAV remote controller. Both plans have their own advantages. The autopilot flight plan is straightforward to collect data. The flight plan is remotely set in the interface and the UAV flies and acquires data automatically. However, the flight plan needs to be adjusted according to the characteristics of the UAV platform, which are maximum flight time, flight speed, height above ground level and horizontal distance.

获取，处理和查看无人机数据

由于研究区域的严峻地形或条件，例如具有陡峭的斜坡或顶部电源线电缆和极线，结构或树木，因此自动驾驶飞行计划可能不合适。在这些情况下，出于安全原因，手动飞行可能会更好。此外，可以使用无人机技术捕获两种不同类型的图像 - Nadir和斜图像（见图1）。倾斜图像增加了三维（3D）模型质量，尤其是在垂直结构上。无人机上的摄像头是收集高质量数据的另一个重要组成部分，相机规范直接影响捕获图像的质量。所得的二维（2D）图像可用于监视道路条件。但是，它们可能不支持准确的测量，因为单个图像没有提供深度信息。

3D模型也可以从2D图像产生。大多数无人机通常包含全球导航卫星系统（GNSS）和惯性测量单元（IMU）传感器，它们为相机位置提供具有厘米级的精度。因此，可以使用Motion（SFM）技术的结构生产3D模型。SFM技术在每个图像中找到可以在连续图像上匹配的绑定点。另外，还使用摄影测量方程来估计摄像头的位置和方向。最后，可以重建感兴趣对象的3D点云。有几种用户友好的商业软件选项（Pix4D Mapper,Agisoft Metashape,3Dsurvey,UASMaster,Photomodeler等）和开源软件（VisualSFM,MicMac,COLMAP））使用SFM技术将2D图像转换为3D点云。此外，可以使用此类软件生产正瘤，数字表面模型（DSM）和数字地形模型（DTM）。可以使用各种软件（例如快速地形建模器和Global Mapper(see Figure 2).

Extracting Road Information from UAV Data

Road information such as road surface, centre line and lane markings, profile, cross-sections and distresses can be extracted from 3D point clouds. These insights into road condition are significant for improving road performance, comfort and safety. In order to collect road information, the road surface first needs to be distinguished from other urban or rural objects and classified. Machine learning algorithms such as Random Forest can be used to classify road surfaces, and these algorithms produce classification results both quickly and with high accuracy. Once the road surface has been classified, other geometric information can be easily extracted.

Road centre line and lane markings are commonly used for road modelling, planning and safety. In addition, this information will be increasingly important for navigation purposes, especially in the context of autonomous driving in the near future. If the road lanes are marked using a specific colour (mostly white or yellow), they can be directly extracted using the RGB feature. RGB values express the colour of the 3D model, and they are transferred from the image processing software using images. In some cases, especially on local roads, road lanes may not be marked in a specific colour or the road lane markings may be damaged and not continuous. In such cases, several methods – such as the improved Voronoi diagram-based algorithms – can be introduced to extract the road centre line and lane markings more robustly and accurately.

‘Road profile’ refers to the vertical section taken along the alignment axis (centre line) of the road. It is important to analyse the incline of the road since this can be a source of danger in the case of icing. Road profiles can be extracted through the DSM. DSM data can be produced from 3D point clouds using various interpolation algorithms. One of the most commonly used interpolation algorithms is the inverse distance weighted (IDW). The height values are recorded as a raster format in the DSM. Then, the Z dimension of the road centre line is extracted from DSM to obtain the profile of the road easily and accurately (see Figure 3).

The road cross-sections provide the creation of a road platform with certain slopes perpendicular to the road centre line. The cross-sections are also important for the transfer of water from the road surface to the roadside and the design of the drainage channels alongside the road. Similarly, the Z dimension of the lines perpendicular to the road centre line is extracted from DSM to obtain the cross-sections of the road easily and accurately (see Figure 3).

Lastly, road distress can also be detected from 3D point clouds. Accurate detection of road distress is an important input for maintenance and repair actions. Timely maintenance and repairs need to be performed to increase the road’s service life, and to maximize comfort and road safety for drivers. In addition, on-time maintenance and repairs may reduce long-term costs. Various methods and techniques are capable of automatically detecting road distress (see Figure 4).

结论

除了在其他学科中使用外，无人机在测量和生产各种最新的道路信息方面都非常重要。特别是，近年来，无人机在道路项目中的使用有所增加，因为它们可以在控制道路库存和安全性，重复的土地调查以及对可持续道路网络的分析以及映射和预测活动中发挥重要作用。自动化信息提取非常有效，可用于小包软件以及传统的GIS软件。结果，可以从基于无人机的数据中提取有关道路构建，库存和道路环境的信息。不可避免的是，在不久的将来，使用无人机系统将确保其作为道路建设中必不可少的测量方法和其他道路信息的生产。

Further reading

SerkanBiçici和Mustafa Zeybek。从无人机衍生的点云中自动提取道路表面困扰的方法。Automation in Construction，122：103475，2021。

SerkanBiçici和Mustafa Zeybek。Effectiveness of training sample and features for random forest on road extraction from unmanned aerial vehicle- based point cloud.Transportation Research记录, 2675(12):401–418, 2021.

Mustafa Zeybek和SerkanBiçici。从UAV的点云数据中提取道路的几何特征。第三国际Confere的诉讼nce on Smart City Applications, pages 435–449. Springer, 2020.

Mustafa Zeybek和SerkanBiçici。使用无人机的道路遇险测量。土耳其遥感和GIS杂志，1（1）：13–23，2020。

Mustafa Zeybek和SerkanBiçici。3D dense reconstruction of road surface from UAV images and comparison of SfM-based software performance.土耳其遥感和GIS杂志, 2(2):96–105, 2021.