In the past few years, there was rapid development in image-based modeling techniques, especially in the structure from motion (SFM) and dense image matching methods. Due to these improvements, cameras are more often applied to capture reality-based 3D models. The photos can be taken either with a static camera (still shots) or a moving camera (video sequence). Video imaging is much easier to apply when compared to still image shooting because the latter requires the proper positioning of the cameras during the capture, and to ensure the required overlapping of the images. However, there are three problems when video image sequences are used for highly detailed modeling. These problems are: the lower resolution of video images, the need of more computational capacity and blur effects due to camera shake on a significant number of images. But if a less accurate model is appropriate then the results would be satisfying. (Barsi, Somogyi, Kapitány 2016)
In this post, you can see an automatically generated model. Thirty-three images (camera: Nikon D5600) from different directions were used. Phase_1 show one of the photos. Then using automatic methods (in this case, it was RealityCapture) common points were detected on the images; you see all identified points with intensity values on Phase_2. Then a 3D mesh was generated on the point cloud in Phase_3. If this wasn’t just for fun, in this phase, manual corrections would be necessary to carry out a more proper surface (e.g., from the directions where some photos had to be eliminated). In Phase_4, you see a textured surface using the intensities on the images.
On the second pic you can see the images and their directions too!
If the images contain an object with known dimensions, the reconstruction could be scaled to the original size. Hence measurements could be done virtually, even if the object itself isn’t easily approachable! For example, with a drone, one could take images about a building/facade, and based on the imagery, a 3D model could be created. Or the same idea could be processed on small scale objects: check out a previous research material regarding the same idea, when the virtual model was also 3D printed!
Link:
https://www.researchgate.net/publication/304003357_INSPECTION_OF_A_MEDIEVAL_WOOD_SCULPTURE_USING_COMPUTER_TOMOGRAPHY
The role of DASHCOAR – What can we offer you?
We analyze objects and building materials, including the inner structure or surface geometry, using various image acquisition techniques.
Why?
- RECONSTRUCT ELEMENTS AND OBJECTS
- ACHIEVE A BETTER UNDERSTANDING OF THE INNER STRUCTURE
- OPTIMIZE MANUFACTURING PROCESS
- DETECTING DEFECTS
We are consulting with construction, and architecture firms to help them apply the right innovative solutions in their daily progress by analyzing their generated DATA for a more efficient and satisfying work environment. We facilitate the building and strengthening of mutually beneficial relations between the municipal and the business sector. Use Big Data analysis to become a part of the digital transformation with us!
#dashcoar #construction #webuildwithdata #datascience #qualityassurance
#innovatewithdata #imageprocessing #drkristofkapitany #alexandrakapitany
/2020.07.12/
Recent Comments