B2-3D++: Uncertainty-aware Hierarchical Registration Network with Domain Alignment
Zhixin Cheng, Jiacheng Deng, Xinjun Li, Li Liu, Xiaotian Yin, Baoqun Yin, Tianzhu Zhang · 2025
The rigid transformation in image-to-point cloud registration is commonly estimated using a coarse-to-fine framework. However, directly and uniformly matching image patches with point cloud patches may lead to focusing on incorrect noise patches during matching while ignoring key ones. Moreover, due to the significant differences between image and point cloud modalities, it may be challenging to bridge the domain gap without specific designs. To address the above issues, we innovatively propose the B2-3Dnet++, which includes the Uncertainty-aware Hierarchical Matching Module (UHMM) and the Adversarial Modal Alignment Module (AMAM). Within the UHMM, we model the uncertainty of critical information in image patches and enable multi-level fusion of image and point cloud features using an uncertainty-aware transformer. In the AMAM, we design an adversarial approach to reduce the domain gap between image and point cloud. Extensive experiments and ablation studies on RGB-D Scene V2 and 7-Scenes benchmarks demonstrate the superiority of our method, making it a stateof-the-art approach for image-to-point cloud registration tasks.