Transferring facade labels between point clouds with semantic octrees while considering change detection
Sophia Schwarz, Tanja Pilz, Olaf Wysocki, Ludwig Hoegner, Uwe Stilla
TL;DR
The paper addresses semantic labeling transfer between two point clouds of the same building by introducing a semantic octree framework coupled with plane-based co-registration via Generalized ICP. It enables automatic label transfer and coarse change detection by representing semantic information and occupancy within octree leaves, with a depth-controlled partitioning to adapt to data density. Empirical results on urban façade data show approximately 82% label-transfer accuracy and a Cohen's kappa around 0.72, with new-point changes around 7.9%, while highlighting limitations related to density variation and complex/rounded features. This approach offers a deterministic alternative to stochastic transfer learning, with potential benefits for data-driven deep learning pipelines and automated multi-temporal semantic labeling of urban scenes.
Abstract
Point clouds and high-resolution 3D data have become increasingly important in various fields, including surveying, construction, and virtual reality. However, simply having this data is not enough; to extract useful information, semantic labeling is crucial. In this context, we propose a method to transfer annotations from a labeled to an unlabeled point cloud using an octree structure. The structure also analyses changes between the point clouds. Our experiments confirm that our method effectively transfers annotations while addressing changes. The primary contribution of this project is the development of the method for automatic label transfer between two different point clouds that represent the same real-world object. The proposed method can be of great importance for data-driven deep learning algorithms as it can also allow circumventing stochastic transfer learning by deterministic label transfer between datasets depicting the same objects.