Afonso Gonçalves webpage

3D Scanning (2c/3c) - 2013/2014 - 1^st Semestre

Ex 1 | 3D Mesh Reconstruction Using Photogrammetry

The objective of this exercise was to explore the use of photogrammetry to survey a physical object and use the data collected to generate a point cloud and subsequent 3D mesh reconstruction. The case study for this project was a small carved stone medallion with relative rough surface and low detail. The method here described has essentially two stages. The first stage is the dense point cloud reconstruction using Visual SFM. The second stage is the 3D mesh reconstruction and color transfer using MeshLab. The process is summarized below:

Stage 1 | Dense Point Cloud Reconstruction using Visual SFM:
1) All the photos were imported into Visual SFM;
2) Started the Computing Missing Pairwise Matching process;
3) Computation of the Sparse 3D Reconstruction and Dense Reconstruction.

Stage 2 | 3D Reconstruction and Texturing Using MeshLab:
1) The point cloud was imported into Meshlab;
2) All spurious points were deleted using the vertex selection tool;
3) The mesh was saved and the ‘Normals’ box checked;
4) The 3D mesh was reconstructed using the Surface Reconstruction: Poisson;
5) Color was transferred from the points to the mesh using Vertex Attribute Transfer;
6) All spurious faces and vertexes were deleted. This step is optional;
7) The final colored mesh was saved.

Links:

Ex 1 Final Report (pdf)
Ex 1 Model File (rar)

Ex 2 | The Terrestrial Laser Scanner approach

The objective of this second exercise was to learn the workflow of aligning point clouds generated by a Terrestrial Laser Scanner. This process is achieved in two stages. The first stage is to down sample each point cloud for easier manipulation. The second stage is to set a point cloud as reference and processing the alignment by choosing homologous points. Afterwards we run the Iterative Closest Points algorithm in order to refine the previous alignment. The aligned point clouds can then be merged together and used to any desired purpose. The process is summarized below:

Stage 1 | Down sampling point clouds:
1) A point cloud is imported into MeshLab;
2) The down sampling is applied using the Poisson-disk Sampling method;
3) The down sampled point cloud is saved;
4) We repeat the previous steps until all point clouds have been down sampled.

Stage 2 | Alignment of point clouds:
1) Using the align tool a point cloud is glued in place and set as reference;
2) Sequentially, we proceed to align every point cloud by choosing homologous points;
3) The ICP parameters are changed and computed in order to refine the alignment;
4) The final aligned composition is saved.

Links:

Ex 2 Final Report (pdf)
Ex 2 Model File (rar)

Ex 3 | Photogrammetric survey of a gothic cloister – Oporto’s Cathedral

This exercise aims to serve as a comprehensive look at the possible uses and applications of photogrammetric survey methods in architecture. Taking a gothic cloister as the case study, the process covers all the steps required for producing useful data that may be used in the architectural context, starting with the photographical survey and acquisition of in-situ data, then approaching the point cloud generation and manipulation and finalizing with the necessary transformations required for properly scaling and orientating f the point cloud. The final product is a collection of information that can be used to extract useful data such as ortho-images, architectural drawings or textured 3D models. The process is summarized below:

Stage 1 | Photographic survey and measurements:
1) Several sets of photos were taken using the SFM (Structure From Motion) method;
2) Additional measurements were collect on site.

Stage 2 | Point cloud generation and editing:
1) The point clouds were generated in VisualSFM from the photos collected;
2) All generated point clouds were imported to MeshLab and spurious points were deleted;
3) The resulting mesh was vertically orientated, rotating it -90º along the X-axis.

Stage 3 | Creating the Target file:
1) Using a CAD program, point coordinates were extracted from the measurements taken on site. The Z-coordinate was discarded;
2) The coordinates were stored in a .txt file and given the name "Target.txt".

Stage 4 | Creating the Source file:
1) The mesh containing the point clouds was imported into CloudCompare;
2) The homologous points to the Target file were selected, following the same order;
3) These points were exported as a .txt file and given the name "Source.txt".

Stage 5 | Calculating the 2D transformation:
1) In JAG3D both "Target.txt" and "Source.txt"were opened in the transformation module;
2) The 2D transformation report was generated and the required values registered;
3) The transformation values were placed in the respective matrix and the final transformation was calculated.

Stage 6 | Applying the transformation:
1) In MeshLab, the project was saved to get the identity matrix;
2) Using a text editor, the values in the identity matrix were replaced with the ones calculated;
3) Applied "Freeze Current Mesh" and saved the mesh.

Links:

Ex 3 Final Report (pdf)
Ex 3 Paper (pdf)
Ex 3 Model File (rar)
Ex 3 Metadata (rar)

↑ Top