Multi-resolution transformer-based point cloud completion network for intelligent manufacture

Due to the limitations of sensor technology and the material (transparency) of the scanned object, the point clouds collected by 3D scanning sensor such as 3D vision sensors will have problems such as missing point clouds and uneven distribution of data points .However, the point cloud complement technique of the past frequently failed to successfully generate the local characteristics of the point cloud due to the point cloud's general structure being refactored. Therefore ,a transformer-based multi-resolution point cloud completion network is suggested. The entire network structure adopts an encoder-decoder structure. Through multi-resolution domain feature embedding, different range of neighborhood features are fused; Then, through the transformer coding structure of adaptive spatial coding, the global features are obtained. Query object adds spatial location information to enhance the absolute spatial location features; Finally, a complete reconstructed point cloud is generated through the transformer decoding structure The experimental results show that compared with PCT, the experimental effect of this paper is improved, and evidence is provided to support the revised algorithm's efficacy.