Nothing Stands Still: A Spatiotemporal Benchmark on 3D Point Cloud Registration Under Large Geometric and Temporal Change

TL;DR

This work introduces Nothing Stands Still (NSS), a spatiotemporal benchmark and dataset for 3D point cloud registration under large geometric and temporal changes in indoor environments. It provides 6 large-area construction-stage scenes captured over months, with defined pairwise and multi-way registration tasks and metrics such as Overlap Ratio ($\text{OR}$) and Temporal Change Ratio ($\text{TCR}$) to quantify data characteristics; a ground-truth ground-truth pipeline uses ICP refinements and cylinder-based context to generate robust pairwise priors. A comprehensive evaluation of state-of-the-art methods (e.g., FPFH, D3Feat, FCGF, Predator, GeoTransformer) shows substantial performance gaps, especially for cross-stage and cross-area settings, underscoring the need for temporally aware and globally consistent registration approaches. The paper also compares NSS to the RIO10 dataset and provides a public evaluation server and leaderboard, highlighting practical impact for robotics, construction progress monitoring, and sustainability in built environments, and pointing to promising directions for future development in spatiotemporal registration.

Abstract

Building 3D geometric maps of man-made spaces is a well-established and active field that is fundamental to computer vision and robotics. However, considering the evolving nature of built environments, it is essential to question the capabilities of current mapping efforts in handling temporal changes. In addition, spatiotemporal mapping holds significant potential for achieving sustainability and circularity goals. Existing mapping approaches focus on small changes, such as object relocation or self-driving car operation; in all cases where the main structure of the scene remains fixed. Consequently, these approaches fail to address more radical changes in the structure of the built environment, such as geometry and topology. To this end, we introduce the Nothing Stands Still (NSS) benchmark, which focuses on the spatiotemporal registration of 3D scenes undergoing large spatial and temporal change, ultimately creating one coherent spatiotemporal map. Specifically, the benchmark involves registering two or more partial 3D point clouds (fragments) from the same scene but captured from different spatiotemporal views. In addition to the standard pairwise registration, we assess the multi-way registration of multiple fragments that belong to any temporal stage. As part of NSS, we introduce a dataset of 3D point clouds recurrently captured in large-scale building indoor environments that are under construction or renovation. The NSS benchmark presents three scenarios of increasing difficulty, to quantify the generalization ability of point cloud registration methods over space (within one building and across buildings) and time. We conduct extensive evaluations of state-of-the-art methods on NSS. The results demonstrate the necessity for novel methods specifically designed to handle large spatiotemporal changes. The homepage of our benchmark is at http://nothing-stands-still.com.

Figures (26)

• Figure 1: Qualitative examples of existing indoor spatiotemporal datasets and Nothing Stands Still (NSS). As shown, existing datasets focus on small and daily changes in living environments, whereas NSS exhibits drastic changes over time. Examples of such changes over different stages are highlighted with a cyan box.
• Figure 2: Overview of the Nothing Stands Still (NSS) benchmark: fragments (${X_A}^t$, ${X_B}^t$, ${X_C}^{t'}$, ${X_D}^{t'}$) captured in a construction site are spatiotemporally registered. First, a pairwise registration step registers individually the pairs of fragments belonging to the same ((${X_A}^t$,${X_B}^t$) and (${X_C}^{t'}$,${X_D}^{t'}$)) or different stages ((${X_A}^t$,${X_C}^{t'}$) and (${X_B}^t$,${X_D}^{t'}$)). Then, a multi-way registration step creates a single and coherent spatiotemporal map of all fragments. Given current methods, this step is initialized by the results of the pairwise one. In this example, we assume overlap occurs for pairs (${X_A}^t$, ${X_B}^t$), (${X_B}^t$, ${X_D}^{t'}$), (${X_C}^{t'}$, ${X_D}^{t'}$), and (${X_C}^{t'}$, ${X_A}^t$). We define the overlapping pairs for entire areas using spatiotemporal graphs, as detailed in Sec. \ref{['sec:multiway_results']}.
• Figure 3: Areas in the Nothing Stands Still dataset at first temporal stage. The building layout and size ranges across areas.
• Figure 4: Sample close-up snapshots of areas in the Nothing Stands Still dataset. Significant changes are occurring per area, starting from an empty scene and reaching the construction of rooms.
• Figure 5: Global registration ground truth, for example, scans in four areas in the Nothing Stands Still dataset. Details about the alignment method for the global ground truth are provided in Sec. \ref{['sec:scan_alignment']}.