ERNet: Efficient Non-Rigid Registration Network for Point Sequences

2025-10-17

AI Summary

To address challenges in non-rigid point cloud sequence registration—including susceptibility to local minima, severe long-range error accumulation, and limited robustness to noise and partial occlusions—this paper proposes a two-stage sliding-window registration framework. The first stage performs coarse inter-frame node estimation via learnable deformation graph prediction; the second stage introduces temporal-aware fine-grained trajectory optimization, explicitly enforcing motion consistency across consecutive frames. The method employs an end-to-end feedforward network, balancing computational efficiency and robustness. Evaluated on DeformingThings4D and D-FAUST benchmarks, our approach achieves significantly higher registration accuracy than state-of-the-art methods, with over 4× faster inference speed. It thus delivers both superior precision and practical applicability for real-world non-rigid registration tasks.

Technology Category

• Computer Vision: Motion & Tracking
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Intelligent Robots: Localization, Mapping, and Navigation

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Registering an object shape to a sequence of point clouds undergoing non-rigid deformation is a long-standing challenge. The key difficulties stem from two factors: (i) the presence of local minima due to the non-convexity of registration objectives, especially under noisy or partial inputs, which hinders accurate and robust deformation estimation, and (ii) error accumulation over long sequences, leading to tracking failures. To address these challenges, we introduce to adopt a scalable data-driven approach and propose ERNet, an efficient feed-forward model trained on large deformation datasets. It is designed to handle noisy and partial inputs while effectively leveraging temporal information for accurate and consistent sequential registration. The key to our design is predicting a sequence of deformation graphs through a two-stage pipeline, which first estimates frame-wise coarse graph nodes for robust initialization, before refining their trajectories over time in a sliding-window fashion. Extensive experiments show that our proposed approach (i) outperforms previous state-of-the-art on both the DeformingThings4D and D-FAUST datasets, and (ii) achieves more than 4x speedup compared to the previous best, offering significant efficiency improvement.

Problem

• Registering shapes to non-rigidly deforming point cloud sequences
• Overcoming local minima and error accumulation in deformation estimation
• Handling noisy partial inputs for accurate sequential registration

Innovation

• Feed-forward model for non-rigid point cloud registration
• Two-stage pipeline predicting deformation graph sequences
• Sliding-window refinement leveraging temporal information