|
Densifying of point cloud from PS-InSAR Method in Urban Area
|
Sajjad Sajedizadeh   , Yasser Maghsoudi Mehrani * |
| K. N. Toosi University of Technology. |
|
|
Abstract: (289 Views) |
| 3D point Clouds have made a significant contribution in airborne and spaceborne observations in recent years. The sensor limitations and processing challenges have been investigated in several studies. We generated 3D point clouds using the persistent scatterers interferometric SAR method in order to estimate obstacles heights. By estimating the height error of the topographic model used in PS-InSAR processing, the permanent scatterer's heights were estimated. The height accuracy was improved by making changes in master image selection. Due to the low density as well as SAR side-looking geometry, the derived 3D point clouds will not represent the complete geometry of complications. In this article, a shape completion deep neural network was trained to increase the point density and complete the shape geometry. By performing learning steps, the network directly maps non-dense and incomplete shape to dense and complete shape geometry. The global features of incomplete shapes were determined while retaining the details to an optimal extent. The loss criterion of this network is based on the Chamfer Distance, which measures the distance between the non-dense input and dense output point clouds. The PS-InSAR processing was done on 27 images of the Sentinel-1 satellite in Philadelphia city, USA. This output obtained from PS-InSAR will be able to estimate the height of various urban complications. We prepared 30000 individual building datasets for training the network for urban areas. The amount of loss obtained in the training process was 0.048 and in the network evaluation process was 0.0482. By evaluating the surface elevation model extracted with lidar reference data of the studied area, the average absolute height estimation error was 4. 67 meters. |
|
| Keywords: 3D Point Cloud Completion, Interferometric SAR, Permanent Scatterers, Elevation Model |
|
|
|
Type of Study: Research |
Subject:
RS
Received: 2022/09/22 | Accepted: 2023/03/6 | ePublished ahead of print: 2024/08/6
|
|
|
|
|
|
|
| Send email to the article author |
|
|
