DTR-Map: A Digital Twin-Enabled Real-Time Mapping System Based on Multi-View Stereo

Digital twins require efficient and cost-effective methods for updating virtual mappings to reflect changes in physical entities. In this paper, we propose DTR-Map, a realtime 3D reconstruction method based on multi-view stereo. Our method utilizes a moving monocular camera as an edge node to collect multi-view data and employs a publisher-subscriber communication framework. By leveraging deep learning-based monocular visual odometry and multi-view stereo matching network, we achieve accurate local camera motion estimation and high-quality output models. Our experimental results demonstrate the method's effectiveness, showcasing its high-quality realtime 3D reconstruction, reconstruction accuracy, and scalability.