DBMovi-GS: Dynamic View Synthesis from Blurry Monocular Video via Sparse-Controlled Gaussian Splatting

2025-06-26

AI Summary

Novel view synthesis from dynamic-blurred monocular video remains challenging due to the reliance of existing methods on static-scene assumptions and high-resolution inputs, resulting in poor robustness under motion blur and non-rigid motion, inaccurate geometry reconstruction, and low visual fidelity. Method: We propose Motion-aware Gaussian Splatting (MAGS), a novel framework that jointly models camera and object motion via dynamic 3D Gaussian representations and motion-aware sparse control. MAGS explicitly decouples and co-optimizes motion-deblurring and geometry reconstruction, eliminating dependence on static priors or high-quality input. Contribution/Results: MAGS achieves stable, high-fidelity novel view synthesis directly from low-resolution, motion-blurred monocular videos. Evaluated on multiple dynamic-blur benchmarks, it significantly improves geometric accuracy (e.g., lower depth RMSE) and visual quality (e.g., higher PSNR, LPIPS) over prior art. MAGS establishes a new state-of-the-art benchmark for blurred monocular novel view synthesis, offering both theoretical novelty—through explicit motion-aware 3D Gaussian optimization—and practical deployability.

Technology Category

• Computer Vision: Motion & Tracking
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Multi-instance/Multi-view Learning

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

Novel view synthesis is a task of generating scenes from unseen perspectives; however, synthesizing dynamic scenes from blurry monocular videos remains an unresolved challenge that has yet to be effectively addressed. Existing novel view synthesis methods are often constrained by their reliance on high-resolution images or strong assumptions about static geometry and rigid scene priors. Consequently, their approaches lack robustness in real-world environments with dynamic object and camera motion, leading to instability and degraded visual fidelity. To address this, we propose Motion-aware Dynamic View Synthesis from Blurry Monocular Video via Sparse-Controlled Gaussian Splatting (DBMovi-GS), a method designed for dynamic view synthesis from blurry monocular videos. Our model generates dense 3D Gaussians, restoring sharpness from blurry videos and reconstructing detailed 3D geometry of the scene affected by dynamic motion variations. Our model achieves robust performance in novel view synthesis under dynamic blurry scenes and sets a new benchmark in realistic novel view synthesis for blurry monocular video inputs.

Problem

• Synthesizing dynamic scenes from blurry monocular videos
• Overcoming reliance on high-resolution or static scene assumptions
• Restoring sharpness and 3D geometry from motion-affected blurry videos

Innovation

• Sparse-controlled Gaussian splatting for dynamic scenes
• Restores sharpness from blurry monocular videos
• Generates dense 3D Gaussians for motion variations