SettleSplat tracks 3D scene changes from single camera views
SettleSplat: Monocular Gaussian Mapping under In-View Scene Changes
Computer Vision and Pattern Recognition
Summary
Tracking how objects move and settle inside a scene seen by just one camera is tricky because some things only move temporarily, while others stay changed. The authors present SettleSplat, a system that keeps a 3D map up to date by telling apart temporary movements from lasting ones using uncertainty measures. This helps in creating consistent 3D representations, even when objects are moved around right in front of the camera. They tested their approach on standard datasets and showed it works better than previous methods that assume scenes stay still.
What this means in practice
- •For robotics developers: Keep robot 3D maps accurate when objects move and settle in view during missions using a single camera.
- •For augmented reality app builders: Create stable AR experiences by updating 3D scenes online even when users move objects in front of the device camera.
Authors
Anh Thuan Tran, Xiyun Song, Yubin Zhou, Heather Yu, Jana Kosecka, Liang Peng
Abstract
Online 3D Gaussian mapping enables fast and high-fidelity novel view synthesis, with broad applications in augmented reality, robotics, and autonomous systems. However, many current methods assume static scenes or filter out dynamics as outliers. In this paper, we study in-view scene changes where objects are manipulated in front of the camera within a single online sequence, unlike prior works that detect changes outside the camera view or between separate captures. This setting entangles transient and persistent dynamics, as objects can be distractors while moving and persistent once they settle. While recent studies start to address distractors by removing them for clean static background, in-view scene changes remain underexplored. To resolve this, we present SettleSplat, a monocular mapping system that maintains up-to-date 3D Gaussian maps under such changes. Specifically, we lift coarse per-pixel uncertainty into Gaussian dynamicity field, then prune transient ones. For persistent changes, we derive geometric contradictions weighted by parameter-free rendered depth uncertainty to separate true changes from under-optimized regions. On Bonn and Co-Fusion, SettleSplat achieves consistent 3D Gaussian maps, outperforming current baselines on both full images and change regions while maintaining comparable speed.