< back

published: 25.06.2025

In modern surveying of power line corridors, accurately classified point clouds are essential. Pointly, in partnership with SWECO, implemented a project to automate the analysis of power line corridors. The project's objective was to develop a custom 3D point cloud classification workflow, built on Pointly's ALS standard classifier and specifically retrained and optimized to the classification requirements of the customer.

The project spanned approximately three months and involved not only the development of the AI model but also the processing of a corresponding dataset of almost 15 billion points. The focus was on the precise automated classification of objects such as power lines, transmission towers, ground, buildings, and vegetation (divided into low, medium, and high). The challenge was to accurately identify and classify these objects within complex, large-scale point clouds, achieving a classification accuracy of at least 95%.

Pointly designed a classification workflow that combines tailored pre-processing with specially optimized post-processing steps. A key enhancement in the post-processing phase involved automated rule-based refinements for specific classes. In particular, a dedicated validation and correction workflow was applied to the Transmission Tower and Power Line classes to ensure accurate separation between the wires and nearby objects.

In addition, adaptive ground filters were used to improve the classification of ground points and reduce false positives. Further accuracy improvements were made by checking for misclassified points on building facades and rooftops, enhancing the recognition of building structures and removing erroneous classifications.

The final workflow achieved over 95 percent accuracy across all classes, reaching up to 99 percent for critical classes such as power lines.

Processing the full dataset required only a few hours on Pointly’s scalable cloud infrastructure.

The collaboration between Pointly and SWECO demonstrates how AI-powered workflows can significantly improve both the quality and efficiency of point cloud processing. The developed classification pipeline includes custom pre-processing, rule-based post-processing and adaptive filtering, specifically addressing the challenges of power line corridor mapping.

The final workflow reached over 95 percent accuracy for all classes and up to 99 percent for the most critical ones. Processing time was reduced to just a few hours using Pointly’s scalable cloud platform. With this optimized solution, SWECO is now equipped to handle similar future projects faster, more reliably, and at reduced cost.

Pointly – Point out what matters.