Learning Compact and Overlap-Biased Interactions for Point Cloud Registration

作     者：Guo, Lin Chen, Zhi Cheng, Senmao Yang, Fan Tao, Wenbing 

作者机构：National Key Laboratory of Science and Technology on Multi-spectral Information Processing School of Artificial Intelligence and Automation Huazhong University of Science and Technology Wuhan430074 China 

出 版 物：《SSRN》 

年 卷 期：2024年

核心收录：

主　　题：Benchmarking 

摘      要：Point cloud registration is a fundamental task in computer vision. Recent Transformer based methods for point cloud registration take advantage of the interaction modeling ability of the attention operation. However, as two of the main challenges of point cloud registration, feature ambiguity, and low overlap are still the bottleneck of the performance in real scenes. In this paper, we present a new neural network to solve these two problems in Transformer architecture. First, we analyze that the dense interaction of the cross-attention (CA) component is a factor for the feature ambiguity problem. Thus, we propose an Optimal Transport guided Cross Attention (OT-CA) to build compact interactions in CA. It uses a Spatial Consistency guided cost Regularization (SCR) to build the cost for the optimal transport problem, and get the weight matrix of CA by solving it. Through the structure information introduced by SCR and the more reasonable interactions in CA, the network can alleviate the feature ambiguity problem with fewer computing resources. Meanwhile, to solve the low-overlap problem, we propose a Separate-and-Joint Overlap Prediction module. It adopts separate branches and training steps for feature matching and overlap prediction to reduce negative impacts between these two tasks, and adopts a joint training process to make full use of overlap information for learning better feature matching. Finally, the proposed modules are embedded into a coarse-to-fine pipeline. Our method shows state-of-the-art performance on three benchmark datasets. © 2024, The Authors. All rights reserved.