PEGS: Physics-Event Enhanced Large Spatiotemporal Motion Reconstruction via 3D Gaussian Splatting

2025-11-21

AI Summary

To address challenges in large-scale spatiotemporal rigid-body motion reconstruction—including limited modeling paradigms, severe motion blur, and insufficient physical consistency—this paper proposes a physics-informed, event-enhanced 3D Gaussian Splatting method. Methodologically, we introduce a triple-supervision framework and a motion-aware simulated annealing strategy to construct the first RGB-Event paired dataset tailored for natural, high-speed rigid-body motion. During reconstruction, we jointly leverage event-stream guidance, acceleration-based physical constraints, and Kalman-filter regularization to co-optimize deblurred geometry and physically plausible motion trajectories. Experiments demonstrate that our approach significantly outperforms state-of-the-art dynamic NeRF and Gaussian-based methods on large-scale dynamic scenes, achieving new SOTA performance in both deblurring accuracy and physical motion consistency.

Technology Category

• Computer Vision: Motion & Tracking
• Intelligent Robots: Motion and Path Planning
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Security and Privacy: Large-scale security measurements

Abstract

Reconstruction of rigid motion over large spatiotemporal scales remains a challenging task due to limitations in modeling paradigms, severe motion blur, and insufficient physical consistency. In this work, we propose PEGS, a framework that integrates Physical priors with Event stream enhancement within a 3D Gaussian Splatting pipeline to perform deblurred target-focused modeling and motion recovery. We introduce a cohesive triple-level supervision scheme that enforces physical plausibility via an acceleration constraint, leverages event streams for high-temporal resolution guidance, and employs a Kalman regularizer to fuse multi-source observations. Furthermore, we design a motion-aware simulated annealing strategy that adaptively schedules the training process based on real-time kinematic states. We also contribute the first RGB-Event paired dataset targeting natural, fast rigid motion across diverse scenarios. Experiments show PEGS's superior performance in reconstructing motion over large spatiotemporal scales compared to mainstream dynamic methods.

Problem

• Reconstructing rigid motion across large spatiotemporal scales with physical consistency
• Addressing severe motion blur through event stream enhancement and deblurring
• Integrating physical priors with multi-source observations for motion recovery

Innovation

• Integrates physical priors with event stream enhancement
• Uses triple-level supervision for physical plausibility
• Employs motion-aware simulated annealing strategy