ActMVS: Active Scene Reconstruction with Monocular Multi-View Stereo
Guo Pu, Yixuan Han, Zhouhui Lian
AI summary
Problem
Existing active reconstruction methods rely on costly depth sensors, while current monocular approaches lack the real-time, globally consistent dense depth required for safe robot and UAV navigation.
Approach
The framework uses a view factor graph with voxel-frame visibility modeling to guide multi-view stereo depth prediction, combined with a global depth optimization algorithm to enforce cross-view consistency for online metric depth estimation.
Key results
- First monocular active reconstruction framework
- View factor graph with voxel-frame visibility modeling
- Global depth optimization for cross-view consistency
- Competitive rendering and mesh accuracy versus RGB-D baselines
Why it matters
Provides a lightweight, vision-only solution for safe autonomous navigation and reconstruction in resource-constrained robots and UAVs.
Abstract
Active scene reconstruction enables robots/UAVs to autonomously plan trajectories and reconstruct environments without costly manual data acquisition. Unlike passive methods, active reconstruction requires real-time construction of high- confidence occupancy maps for collision-free navigation. Ex- isting approaches rely on depth sensors for occupancy map updates, increasing platform cost and weight. To advance spatial intelligence, we aim for a vision-only monocular solution. However, current monocular scene reconstruction methods operate offline and fail to deliver globally consistent dense depth at the frame rates required for robots/UAVs navigation. To bridge this gap, we introduce ActMVS, the first framework for monocular active reconstruction. Our framework integrates a view factor graph construction for informed Multi-View Stereo depth prediction, along with a global depth optimization, to enable the online generation of high-quality, globally consistent dense depth maps. This enables monocular robots/UAVs to maintain reliable occupancy maps for safe trajectory plan- ning during reconstruction. Experiments on Replica datasets demonstrate performance competitive with RGB-D methods. Our code and data are available at https://github.com/ TrickyGo/ActMVS.