Multi-modal Fusion and Cross-Attention Module for Point Cloud Registration
2024 IEEE 6th Advanced Information Management, Communicates, Electronic and Automation Control Conference (IMCEC)(2024)
摘要
The registration of point clouds is a crucial technology in the fields of computer vision and image processing. However, existing point descriptors for point cloud registration solely rely on structural information, neglecting the texture information inherent in the images themselves. The addition of texture features to the point descriptor is therefore considered in order to enhance its distinctiveness. Specifically, we introduce a Point Cloud Structure Extractor (PSE) to extract structural information and employ a cross-attention module to capture weighted texture information. Subsequently, we fuse these two types of information to generate multi-modal point descriptors. Our method has been trained and tested on 3DMatch and KITTI datasets, demonstrating its superior accuracy compared to state-of-the-art approaches.
更多查看译文
关键词
point cloud registration,point descriptor,3D reconstruction,cross-attention module
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要