以光速处理激光雷达

Author: Linda Duffy

LiDAR数据可能是一件好事。最新一代的机载激光雷达传感器能够收集非常密集的点云，其中包含许多应用程序的关键信息；但是，对于缺乏投资所需计算基础架构的资源的初学者和公司来说，处理和管理这些庞大的数据集令人生畏。较长的处理时间可以缩短整个工作流程并延迟最重要的步骤 - 分析数据并向客户提供可行的情报以实现决策。

空中激光雷达数据

SolSpecis a geospatial data collection processing and analytics company with unique expertise in remote sensing applications of earth science. This Denver-based company started out offering drone acquisition and processing services, primarily to the oil and gas industry, but soon pivoted to a scalable aerial strategy serving a broad range of customers.

In 2019, SolSpec purchased a pair ofLeica TerrainMapperlinear-mode LiDAR sensors with built-in digital cameras that capture 4-band data (RGB and NIR) and cover more ground in a far shorter time than a drone. The aircraft-mounted system flies much higher and faster, efficiently collecting a broader area, and making it a scalable tool for diverse situations. On average, a single day of capture at 20 points per square metre (ppsm) produces 1 TB of highly accurate data.

“By collecting higher point densities with the TerrainMapper, we provide better data to our customers,” says Seth Tribbey, VP of Sales and Marketing at SolSpec. “This translates into more valid analytics to support the decisions being made.”

高性能计算技术

After identifying the traditional processing workflow as an obstacle for expedited delivery of information based on aerial LiDAR data, SolSpec realised the cloud offered the opportunity to scale up multiple machines to process data in parallel. To harness the near-infinite computing power of the cloud, the innovative company built the SolSpec Engine on a high-performance computing and machine learning backbone.

Leica HxMap, a multi-sensor platform designed to leverage the benefits of the TerrainMapper data, has an open interface processing workflow that accommodates cloud processing, which is unique in the industry. HxMap can be installed from a single workstation to high end cluster environment with hundreds of nodes for high volume production. The workflow can be orchestrated in different engines using the many command line applications that can be scripted to fulfill various workflows to produce different output products.

When deployed on top of the SolSpec Engine in the Amazon Web Services (AWS) cloud, HxMap processes large volumes of aerial data in a compressed time frame. Deployment of HxMap into the AWS cloud on Linux allowed SolSpec to rapidly scale data processing capacity. The company increased its processing speed for large airborne LiDAR data sets by up to 60 times, which reduced post-processing costs and decreased the data delivery time to the customer. In addition, by building decision-ready analytics into the SolSpec Engine, SolSpec significantly reduced “time-to-decision” with LiDAR data.

“The flexibility that Hexagon has shown in embracing new technology and the willingness to work with us were main drivers for our partnership,” Tribbey says. “We’ve automated the proprietary TerrainMapper format, taken core libraries and adapted them to run on top of a cloud engine. This structure allows us to go from raw wave-form data to a point cloud to a terrain model in a fraction of the time compared to conventional providers.”

SolSpec Geohazard Analytic is built on machine-learning and trained on 15,000 historic landslides to identify landslide signatures in a digital terrain model even in areas covered by vegetation. Courtesy SolSpec.

Real-life applications

One flagship product for SolSpec is Right-of-Way (ROW) Integrity Management. Pipeline operators frequently deal with environmental conditions like landslides and erosion that threaten to rupture pipes and interrupt service. If there is a thick tree canopy obscuring the slide activity, it is difficult to identify problems. The TerrainMapper effectively penetrates tree canopies to collect high density point clouds that contain a great level of detail. After processing data in the cloud, machine-learning algorithms are used to identify hazards through the canopy, alerting customers more quickly to potential issues.

同样，SOLSPEC的植被管理解决方案高度依赖于富的高密度点云以及TerrainMapper的高分辨率4波段光谱数据。通过利用光谱和结构数据，SOLSPEC数据科学家开发了算法，这些算法检测植被侵占电力线的位置，预测野火风险，识别湿地，对植被类型进行分类用于评估开垦活动以及许多其他应用。

“We have a perfect marriage between the data collection with the TerrainMapper and the processing with HxMap,” says Tribbey. “This allows us to leverage the nearly unlimited computing power on AWS and deliver information incredibly faster.”

SOLSPEC的表面水文分析模型的水文能量和高分辨率数字地形模型的表面流量准确鉴定侵蚀问题。礼貌的索尔斯克。

Streamlining the LiDAR value chain

SolSpec’s ultimate goal is to transform aerial data into actionable risk analysis, decision support and predictive modelling. By combining aerial data and project-specific data with machine learning tools, SolSpec produces comprehensive reports about site conditions more quickly than other providers.

Approximately half of a LiDAR project’s cost is acquiring data, and the other half is processing. A conventional provider may take 30 hours to process 3 square miles of data (20 ppsm, QL1) in the office, while the SolSpec Engine does the same work in 20 minutes in the cloud. Processing with HxMap on the super-fast SolSpec Engine is a key competitive advantage to support time-sensitive decision making.

“Hexagon was an obvious choice for us as a partner because of its reputation in the aerial LiDAR industry,” continues Tribbey. “Its sensors and software are very reliable and produce excellent results.”

在云中加工的综合时间节省和使用机器学习算法自动化分析的时间增加了LiDAR数据的价值，并将其实用性扩展到各个市场。

Tribbey解释说：“ SOLSPEC帮助其客户识别，测量，优先级和监视对其基础设施最常见的环境威胁。”“我们通过更快地处理，管理数据，运行分析并提供客户需要的答案来简化整个LIDAR值链。这使我们的客户能够迅速做出有关侵蚀，地球岩，植被，侵占，结构等的决定。”