Warning

This page is located in archive.

The goal of exercises is to practice the 3D reconstruction techniques explained at lectures. During the exercises, students will build up complete system for reconstruction of a surface of a 3D scene given its images. Implementation if this system and its use for the reconstruction is solved as a term project, and it is subject of the exercises. Students are working individually during the exercises as well as on the term project.

Whole system is build from several building blocks − elementary methods. These methods will be explained in the lectures gradually. Students will implement some methods, another methods are already available and the task is to integrate them into the system.

Whole problem is separated into following four phases. These consist of methods that are related and possibly cooperates. Each phase represents relatively independent part, and it covers several week classes. There are defined required results for each phase. Whole project is finished by submission of a final results of reconstruction.

- Input data capture and preprocessing
- Epipolar geometry on image pairs.
- Multiple cameras and structure of the scene.
- Surface reconstruction.

The term project and other tasks should be solved using the Matlab environment. There are some methods, that students need not implement ^{1)}.
These methods can be obtained from the code repository.
These codes can be subject of small evolution and bug-fixing. It is thus recommended always to use the latest revision. In case that you found a bug, please announce this using the forum.

During the implementation, students should build a set (library, toolbox) of general elementary tools and methods. The functions of this Toolbox of Elementary and Helper Functions will be specified during the term. This toolbox should be submitted (repeatedly, as the number of functions will grow) to the upload system for the automatic check of validity (`_toolbox`

assignment).

**Important:** Without prior permission from the teacher, it is **not allowed to use different software tools** than the ones specified in the description. However, this does not apply for visualization and presentation of results (vrml and so).

Week | Date | Phase | Details | Submission | Points |
---|---|---|---|---|---|

1. | 24.9. | 1. Input data | Introduction, term project specification, capture, camera calibration. | ||

0. Basic geometry, Matlab | Points and lines in a plane. | ||||

2. | 1.10. | Perspective camera. | Task 0-1 | 1 | |

3. | 8.10. | Robust maximum likelihood estimation of a planar line. | Task 0-2 | 2 | |

4. | 15.10. | 2. Seeking of Sparse Correspondences | WBS matcher, sparse correspondences. | Task 0-3 | 3 |

5. | 22.10. | 0. Basic geometry (continued) | Estimation of two homographies | Task 2 | 3 |

6. | 29.10. | 3. Robust estimation of calibrated epipolar geometry of image pairs | Task 0-4 | 4 | |

7. | 5.11. | –continued | |||

8. | 12.11. | 4. Multiple cameras and structure of the scene | Calibration of poses of a set of cameras | ||

9. | 19.11. | Sparse point cloud reconstruction | Task3 | 10 | |

10. | 26.11. | Test 1 | |||

11. | 3.12. | Optimisation of points and cameras by bundle adjustment | |||

12. | 10.12. | 5. Surface reconstruction | Epipolar rectification and dense matching. Dense point cloud reconstruction. Final 3D surface reconstruction. | Task 4 | 14 |

13. | 17.12. | ||||

14. | 7.1. | Consultation, reserve, credit (if eligible) | Task 5 (12.1.2019) | 8 |

There is 45 pts assigned for a continuous work during exercises.

Small problems and the term project are solved during the labs. Successfully solved and demonstrated
(to the teacher) or submitted (to the upload system where required) problem is credited by appropriate number of points. All problems and project sub-tasks have deadlines. **Later demonstration/submission will be penalized by -15% of the nominal points for the task if one week later or by -30% if two and more weeks later.**

Final results of your work should be submitted to the Upload system. Upload the following parts:

- three vrml models (in archive, assignment
`vrml`

),- cameras and sparse point cloud,
- dense point cloud,
- surface,

- your codes (in archive, assignment
`codes`

).

Students who wants to obtain the credit in the last week of the term should submit the results in time. In the case of later submission, the credit will be granted after consultation with the teacher.

but of course they can, in the interest of intimate understanding

courses/tdv/labs/start.txt · Last modified: 2019/09/19 08:42 by xmatousm