Temporal comparison of construction sites using photogrammetric point cloud sequences and robust phase correlation
Registration of multi-temporal data is an important task when conducting construction monitoring and analysis using 3D point clouds acquired at different time points. However, due to the complexity of scenes in construction sites and the intrinsic attributes of the datasets (e.g., noise, outliers, a...
Saved in:
| Published in: | Automation in construction 2020-09, Vol.117, p.103247, Article 103247 |
|---|---|
| Main Authors: | , , , |
| Format: | Article |
| Language: | eng |
| Subjects: | |
| Online Access: | Get full text |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Registration of multi-temporal data is an important task when conducting construction monitoring and analysis using 3D point clouds acquired at different time points. However, due to the complexity of scenes in construction sites and the intrinsic attributes of the datasets (e.g., noise, outliers, and uneven densities), the registration of multi-temporal datasets is a challenging task. To this end, in this paper, we propose a fast and marker-free method for coarse registration of multi-temporal point clouds by integrating a projection-based strategy with Fourier-based signal matching methods, consisting of three major steps: decoupling of 3D transformation with projection-based dimensionality reduction, estimation of horizontal transformation by matching 2D horizontally projected images, and estimation of vertical translation by matching 1D vertical signals. In the decoupling step, the principal projection plane is firstly identified through a robust plane fitting method. Then, each point cloud is decomposed into two independent parts according to the estimated projection plane, namely a 2D image and a 1D signal. The 2D image is obtained by mapping 3D points to the principal projection plane along the vertical direction. Correspondingly, the 1D signals are attained through projecting 3D points to the principal axis in the horizontal direction. In the following step, the rotation, scaling, and translation parameters between 2D images of different point clouds are decoupled with Fourier-Mellin transform. These transformation parameters are then solved with a robust 2D phase correlation algorithm. Simultaneously, 1D signals of non-registered point clouds are matched with a robust 1D phase correlation approach. Finally, combining estimated transformation parameters from both 2D images and 1D signals, the multi-temporal point clouds can be efficiently registered. Our registration method is tested using one TLS benchmark dataset and two multi-temporal point cloud sequences from different construction sites to validate the versatility and effectiveness of the proposed workflow. In terms of the registration accuracy, our proposed method can achieve satisfying performance, with accurate registration for both rotation and translation estimation.
•A novel coarse registration framework with low-frequency components is proposed.•A concept converting 3D registration to the robust matching of 3D signals is given.•A projection-based dimensionality reduction is designed to |
|---|---|
| ISSN: | 0926-5805 1872-7891 |