Point Cloud Registration With Object-Centric Alignment

Point cloud registration is a core task in 3D perception, which aims to align two point clouds. Moreover, the of clouds with low overlap represents harder challenge, where previous methods tend fail. Recent deep learning-based approaches attempt overcome this issue by learning find overlapping regions whole scene. However, they still lack robustness and accuracy, thus might not be suitable for real-world applications. Therefore, we present novel pipeline that focuses on object-level alignment provide robust accurate By extracting completing missing points object interest, rough can achieved even captured from widely apart viewpoints. We quantitative qualitative evaluation synthetic data Kinect v2. The proposed approach outperforms current state-of-the-art more than 29% w.r.t. recall introduced dataset. show overall performance increases due alignment, while baselines perform poorly as take entire scene into account.