OpenTrench3D: A Photogrammetric 3D Point Cloud Dataset for Semantic Segmentation of Underground Utilities

TL;DR

OpenTrench3D addresses the lack of public 3D datasets for underground utilities by introducing a smartphone photogrammetry–derived point-cloud dataset with 310 annotated clouds across water and district-heating areas. The authors benchmark three state-of-the-art 3D segmentation networks (PointNeXt, PointVector, PointMetaBase) and demonstrate cross-domain transfer learning by pretraining on water-area data and fine-tuning on heating-area data, reporting $mIoU$ and $mAcc$ gains. They propose a utility-owner-centric five-class scheme (Trench, Main Utility, Other Utility, Inactive Utility, Misc) and provide dataset statistics, capture workflow, and evaluation protocols. The work shows that transfer learning can effectively extend segmentation to new utility types with limited labeled data, enabling cost-efficient underground utility surveying and mapping and potentially reducing excavation damages.

Abstract

Identifying and classifying underground utilities is an important task for efficient and effective urban planning and infrastructure maintenance. We present OpenTrench3D, a novel and comprehensive 3D Semantic Segmentation point cloud dataset, designed to advance research and development in underground utility surveying and mapping. OpenTrench3D covers a completely novel domain for public 3D point cloud datasets and is unique in its focus, scope, and cost-effective capturing method. The dataset consists of 310 point clouds collected across 7 distinct areas. These include 5 water utility areas and 2 district heating utility areas. The inclusion of different geographical areas and main utilities (water and district heating utilities) makes OpenTrench3D particularly valuable for inter-domain transfer learning experiments. We provide benchmark results for the dataset using three state-of-the-art semantic segmentation models, PointNeXt, PointVector and PointMetaBase. Benchmarks are conducted by training on data from water areas, fine-tuning on district heating area 1 and evaluating on district heating area 2. The dataset is publicly available. With OpenTrench3D, we seek to foster innovation and progress in the field of 3D semantic segmentation in applications related to detection and documentation of underground utilities as well as in transfer learning methods in general.

Figures (8)

• Figure 1: Point clouds from OpenTrench3D: (a) In the water project areas the Main Utility class is made up of water utility pipes. (b) In the district heating project areas the Main Utility class is made up of district heating utility pipes. Find description of dataset and classes in section \ref{['sec:dataset']}.
• Figure 2: Map overview of the five water areas (top left and middle) and the two district heating areas (right), with zoom-in maps of selected parts of some areas. Two point cloud samples are presented from the water and heating areas, respectively, and utilities from the Main Utility, Other Utility, and Inactive Utility classes are highlighted for comparison.
• Figure 3: An overview of the dataset sub-sets of OpenTrench3D used in the 5-fold cross-validation on water areas and fine-tuning evaluation on heating areas as described in section \ref{['subsec:experiment-desc']}.
• Figure 4: Results from the fine-tuning experiments described in section \ref{['subsec:description-dataset']}. For comparison, we display results from pre-trained models (red square) and results of models solely trained on samples from Heating Area 1 (Baseline).
• Figure 5: Qualitative results of pre-trained PointNeXt model, PointNeXt models trained on 1 and 10 samples from Heat Area 1 and finally pre-trained and fine-tuned PointNeXt models fine-tuned on 1 and 10 samples were only weights of the segmentation head are tuned. A set of additional examples are seen in the supplementary material.