Interactive 3D Modelling
3D reconstruction and modeling techniques based on computer vision show a significant improvement in recent decades. Despite the great variety, a majority of these techniques depend on specific photographic collections or video footage. For example, most are designed for large data collections, overlapping photos, captures from turntables or photos with lots of detectable features such as edges. If the input, however, does not fit the particular specification, most techniques can no longer create reasonable 3D reconstructions. The purpose of this project is the development of a software components that overcome these drawbacks in cooperation with the user's real world knowledge. Upon this basis, this project should lead to an interactive 3D reconstruction and modeling software
- to achieved accurate, reliable, non-monolithic, functioning and CAD-ready models
- in cooperation with an non-expert user
- from video footage
- in a few simple user interactions, only.
The the main drive of this project is the multitude of computer vision based techniques, that reconstruct and model 3D objects or scenes from photographs or video footage captured in 2D (monocular) or 3D (stereo). Many of these techniques and approaches reconstruct objects or scenes with automatic algorithms from image sequences. But would an architect or an engineer call the resulting 3D reconstruction a CAD-model? Is a non-expert user able to apply these tools? Can the user apply the method without special hardware like a rotating plate, laser, stereo camera or even a main frame computer? Are users able to reconstruct models of real world objects in such a way that they will be able to translate them back to real world replicas? The answer to these questions is mostly – No – but why?
Therefore, this project reviews automatic and interactive 3D reconstruction and modeling techniques, compare them and and analyze their weaknesses. Under these perspectives, it should be discussed how an interactive computer vision application could be employed to overcome existing weaknesses. Finally, an application should be developed for the interactive creation of non-monolithic, functioning 3D models — models that architects or engineers would call a model, and CAD-ready models that can be applied for simulation tasks, reverse engineering, replication and many more purposes.
Osnabrück — Synthetic Scalable Cube Dataset - SCIA'17
Multimedia Container Formats for Instantaneous Data Visualizations - JEMR, ETVIS'16, WACV'17, VISAPP'17
Interactive feature growing - EPIC@ECCV'16
iSeg - semi-automatic ground truth annotation tool - K-CAP'15; ICPRAM'16
Bio-Inspired Architecture for Deriving 3D Models from Video Sequences - ACCV'16; ICIAP'17
Evaluation of multi-view 3D reconstruction software - CAIP'15