Point Cloud Registration Algorithm Based on Offset Cross Attention
Point cloud registration plays an important role in the development of machine vision,artificial intelligence and other fields.A point cloud registration network(OCADGCNN)based on offset cross-attention is presented to overcome the low accuracy and poor robustness of traditional point cloud registration algorithms and existing deep learning point cloud registration algorithms.The offset attention module is in-serted into the Dynamic Graph Convolution Neural Network(DGCNN)to extract global eigenvectors,which can make full use of the local structure and spatial semantics information of point clouds and reduce the loss of information.Include residual connections in feature extraction to improve network performance.The interactive attention module is used to exchange information between global features,enhance related in-formation,and suppress the interference of non-overlapping area information.The experimental results show that the registration effect of OCADGCNN is better than ICP,PointNetLK,PCRNet,OMNet and DOPNet in both noise-free and low-noise ModleNet40 data sets,and the registration accuracy is high.In the experiments of unknown categories,the model has high generalization ability and good versatility,and can better handle low overlap point clouds when the integrity of point clouds is reduced.
point cloudregistrationdeep learningattention mechanismdynamic graph convolutionfeature interaction
万方数据
维普
