3D Scene Change Modeling With Consistent Multi-View Aggregation

2025-12-28

AI Summary

Existing 3D change detection methods suffer from spatial inconsistency and ambiguity in distinguishing pre- and post-change states. To address these issues, we propose SCaR-3D, the first framework leveraging Signed Distance Field (SDF)-guided 2D differential modeling and multi-view consistency voting for object-level, direction-aware change localization. We further introduce a dynamic region selection and update strategy to enable continual scene reconstruction. Additionally, we construct CCS3D—the first controllable, compositional synthetic dataset for 3D change detection. Our method integrates multi-view 3D Gaussian Splatting (3DGS) reconstruction, SDF-based differential representation, voting-based aggregation, and incremental optimization. On CCS3D, SCaR-3D significantly outperforms state-of-the-art methods: it achieves a +12.6% mIoU gain, 2.3× faster inference, and simultaneously ensures high accuracy, computational efficiency, and structural fidelity.

Technology Category

• Computer Vision: 3D Computer Vision
• Knowledge Representation and Reasoning: Action, Change, and Causality
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Change detection plays a vital role in scene monitoring, exploration, and continual reconstruction. Existing 3D change detection methods often exhibit spatial inconsistency in the detected changes and fail to explicitly separate pre- and post-change states. To address these limitations, we propose SCaR-3D, a novel 3D scene change detection framework that identifies object-level changes from a dense-view pre-change image sequence and sparse-view post-change images. Our approach consists of a signed-distance-based 2D differencing module followed by multi-view aggregation with voting and pruning, leveraging the consistent nature of 3DGS to robustly separate pre- and post-change states. We further develop a continual scene reconstruction strategy that selectively updates dynamic regions while preserving the unchanged areas. We also contribute CCS3D, a challenging synthetic dataset that allows flexible combinations of 3D change types to support controlled evaluations. Extensive experiments demonstrate that our method achieves both high accuracy and efficiency, outperforming existing methods.

Problem

• Detects object-level changes from dense and sparse image sequences
• Separates pre- and post-change states using consistent multi-view aggregation
• Enables continual reconstruction by selectively updating dynamic scene regions

Innovation

• Multi-view aggregation with voting and pruning
• Signed-distance-based 2D differencing module
• Selective continual reconstruction of dynamic regions