Abstracted Workflow Framework with a Structure from Motion Application
Saved in:
| Title: | Abstracted Workflow Framework with a Structure from Motion Application |
|---|---|
| Authors: | Rossi, Adam J |
| Source: | Theses |
| Publisher Information: | RIT Digital Institutional Repository |
| Publication Year: | 2014 |
| Collection: | Rochester Institute of Technology: RIT Scholar Works |
| Subject Terms: | 3D modeling, 3D reconstruction, Computer vision, Image chain, Software workflow, Structure-from-motion |
| Description: | In scientific and engineering disciplines, from academia to industry, there is an increasing need for the development of custom software to perform experiments, construct systems, and develop products. The natural mindset initially is to shortcut and bypass all overhead and process rigor in order to obtain an immediate result for the problem at hand, with the misconception that the software will simply be thrown away at the end. In a majority of the cases, it turns out the software persists for many years, and likely ends up in production systems for which it was not initially intended. In the current study, a framework that can be used in both industry and academic applications mitigates underlying problems associated with developing scientific and engineering software. This results in software that is much more maintainable, documented, and usable by others, specifically allowing new users to extend capabilities of components already implemented in the framework. There is a multi-disciplinary need in the fields of imaging science, computer science, and software engineering for a unified implementation model, which motivates the development of an abstracted software framework. Structure from motion (SfM) has been identified as one use case where the abstracted workflow framework can improve research efficiencies and eliminate implementation redundancies in scientific fields. The SfM process begins by obtaining 2D images of a scene from different perspectives. Features from the images are extracted and correspondences are established. This provides a sufficient amount of information to initialize the problem for fully automated processing. Transformations are established between views, and 3D points are established via triangulation algorithms. The parameters for the camera models for all views / images are solved through bundle adjustment, establishing a highly consistent point cloud. The initial sparse point cloud and camera matrices are used to generate a dense point cloud through patch based techniques or ... |
| Document Type: | text |
| File Description: | application/pdf |
| Language: | unknown |
| Relation: | https://repository.rit.edu/theses/7814; https://repository.rit.edu/context/theses/article/8821/viewcontent/ARossiThesis5_2014.pdf |
| Availability: | https://repository.rit.edu/theses/7814 https://repository.rit.edu/context/theses/article/8821/viewcontent/ARossiThesis5_2014.pdf |
| Accession Number: | edsbas.AD0D0A26 |
| Database: | BASE |
Nájsť tento článok vo Web of Science Be the first to leave a comment!