PBFormer: Point and Bi-Spatiotemporal Transformer for Pointwise Change Detection of 3D Urban Point Clouds

Change detection (CD) is a technique widely used in remote sensing for identifying the differences between data acquired at different times. Most existing 3D CD approaches voxelize point clouds into grids, project them 2D images, or rasterize digital surface models due to irregular format of and variety changes three-dimensional (3D) objects. However, details geometric structure spatiotemporal sequence information may not be fully utilized. In this article, we propose PBFormer, transformer network with Siamese architecture, directly inferring pointwise bi-temporal clouds. First, extract sequences from using k-nearest neighbor method. Second, uniquely use as an encoder feature bitemporal Then, design module fusing features effectively detect change features. Finally, multilayer perceptrons are obtain results. Extensive experiments conducted on Urb3DCD benchmark show that PBFormer outperforms other excellent cloud tasks.