基于扩散模型的条件聚合三维点云重建方法  

A Conditional Aggregation 3D Point Cloud ReconstructionMethod Based on Diffusion Model

作　　者：冯泽 章建峰 韩光照 王兴梅[1] FENG Ze;ZHANG Jian-feng;HAN Guang-zhao;WANG Xing-mei(College of Computer Science and Technology,Harbin Engineering University,Harbin 150001,China;Shanghai Marine Equipment Research Institute,Shanghai 200031,China;Harbin Electric corporation Marine Intelligent Equipment Company Limited,Harbin 150028,China)

机构地区：[1]哈尔滨工程大学计算机科学与技术学院,哈尔滨150001 [2]上海船舶设备研究所,上海200031 [3]哈尔滨电气集团海洋智能装备有限公司,哈尔滨150028

出　　处：《航空电子技术》2024年第2期48-55,共8页Avionics Technology

基　　金：国家级重点实验室开放基金项目(KY10600220048)。

摘　　要：随着计算机视觉技术的发展,利用图像进行三维点云重建已成为连接现实世界与数字世界的关键技术,现有的研究更多地应用全局图像特征推断物体的合理形状和外观,导致重建的三维点云细节缺失。为解决这一问题,在扩散模型基础上,引入条件聚合方法融合图像信息作为扩散条件,构建条件扩散的三维点云重建网络模型,利用图像特征提取模块捕捉图像特征,充分挖掘图像信息,采用投影变换聚合点云与对应的局部图像信息,使扩散模型关注点云中每一个点的重建过程,提高三维点云重建的精度和细节表现。在ShapeNet数据集上的对比和消融实验表明,该模型能够充分利用图像信息,增强重建结果的细节表现力,提高重建性能。With the development of computer vision,the use of images for 3D point cloud reconstruction has become a key technology to connect the real physical world and the digital world.Existing research often applies more global image features to infer objects with reasonable shapes and appearances,resulting in reconstruction The 3D point cloud has the problem of missing details.In order to solve this problem,based on the diffusion model,aconditional aggregation method is introduced to fuse image information as diffusion conditions,and a conditional diffusion three-dimensional point cloud reconstruction network model(CDPRN)is constructed,and the image feature extraction module is used to capture Image features,fully mining image information,using projection transformation to aggregate point clouds and corresponding local image information,allowing the dffusion model to focus on the reconstruction process of each point in the point cloud,improving the accuracy and detail performance of three-dimensional point cloud reconstruction.Comparison and ablation experiments on the ShapeNet data set show that CDPRN can make full use of image information,enhance the detail expression of the reconstruction results,and improve reconstruction performance.

关 键 词：扩散模型 三维点云重建 条件聚合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]