I am trying to get 3D images of the corroded reinforcement steel bars from 2D images taken at 360 degrees. I need to measure the cross-section at different location using some software which can measure the dimensions from 3D images. I am looking for a 0.01 mm resolution for the measurement and an accuracy for the 3D image that can capture even the small dents in the corroded steel rebar. Could you please suggest any software which can serve this purpose.
I'm not sure if I understand well what do you wanna, but one way to do that is to use ImageJ to get a histogram of the 2D image and the plot a 3D chart with Origin and even to try a 3D image of that histogram. ImageJ is a good code to solve this kind of problem. Good luck.
VisualSFM may help you reconstruct a 3D model from a number of 2D pictures.
http://ccwu.me/vsfm/
I have gained good experiences with the PhotoModeler Software. With a series of images taken at 360 degrees you should be able to get very accurate 3D models and if you use a well calibrated camera you are able to set the scale for your model with no problems.
Hope this helps and good luck!
There are several different software tools which are able to do the task. You just have to google "structure from motion software" or "multiple view geometry software". As already mentioned, VisualSFM is one solution, another would be to e.g. use Photosynth in combination with Meshlab.
The accuracy you get not only depends on the software, but even more on the quality and amount of images you take as input. A 0.01mm accuracy is exrremly challenging if not impossible.
Reconstruct on 3D model from 360 degree 2D pictures is doable with several software such as Photoscan from AgiSoft, or 123D catch from Autodesk (http://www.123dapp.com/catch), you have also ReCap from autoDesk (https://recap360.autodesk.com/Signin.aspx#home). But they all suffer the same issue. Non more than mm accuracy maximum. Even with very good quality pictures. your level of accuracy seem quite high even for CT-scan as a CT-scan is about 0.5 mm.
ImageJ might help you as it is a scientific software. I know this software can reconstruct 3D object from CT-scan but I do not know the capabilities to handle 360 degree 2D picture.
I would like not to be pessimist but hand measurement could be the faster and the more accurate way for measurements about 0.01 mm precision. Or may be laser scanning but it is expensive.
If you find a software able to do what you want, please let us know.
Best luck
Geoffrey
A possible solution ( a conjecture open to correction ) with hopefully minimum technicalities..
1. Calibrate the camera with known flat strips ie distance between pixel points and
actual distance in object. (perhaps some elementary lens optics.
2 increase the number of images around 360deg so the circle(rod) can be
approx by rectangular strips , then the rod could be built up as rectangles stitched together.
3.In the present case a distinct mark on each strip of the rod , could be used
to synch the images..
Good luck
Cheers
Dear Ranjhita,
For this answer I assume that you don't have to achieve your goal once from a given set of images, but instead are designing a method to measure such corrosion routinely.
In that case: Use a time of flight camera and image at several rotation angles. There are now industrial tim of flight cameras that can capture a full depth profile image-plane as a snapshot that e.g. allows measurement of a human fingerprint. This approach probably would bring you quicker into the desired range of precision.
Kind regards and good luck, Fokko Wieringa
PS: In case you must use normal camera data, I also uploaded a publication on 3D reconstruction from video stream in the medical field.
In principle Pix4D (https://pix4d.com/) would be a good software choice, though your resolution/accuracy requirements are very high. If the photos are very detailed you should be able to generate a suitable 3D model - geolocating to an absolute reference framework will be a greater challenge.
There are many software that are designed for "close rage photogrammetry", which is the topic you should look into. I personally worked with Agisoft PhotoScan and it was very satisfactory.
However, about the measurement resolution, one of the important parameters is the image acquisition resolution. It means you need a camera body with high megapixels, and camera lens with enough magnification.
Good luck.
It is a matter of fact if you take macro photography you might reach 0.01 mm accuracy/precision. Usually the accuracy/precision of photogrammetry is about 1%. However, you will to deal with depth of field with macro photography. So not sure how the software will deal with that as the background will be very blurry. But you can still try. After all, you have a fully functional demo version for 1 month available ont he website of Agisoft Photoscan.
Best regards
Some additional aspects that might be important..
1. Source of illumination .viz monochromatic or not and angle of illumination..
2. It may require post processing to remove artifacts such as reflectance . enhancement of Texture, and distortion in color so need to convert to HSV etc...
Cheers
Hi Ranjitha,
As Geoffrey Marchal said, the accuracy of 3D model is mostly depending on the resolution of images and the overlapping area between pairs of images. From my previous work, the relative distance error between two points is around 1% by processing high overlapping UAV sequences, using the open source code "Bundler". A GPU accelerated VisualSFM software developed by Wu Changchang is free to the public, which is also a very good 3D reconstruction software.
Hope be helpful
Zhihua
Stereo Vision - MATLAB
Computer Vision System Toolbox.
The 3D information can be obtained from a pair of images, also known as a stereo pair, by estimating the relative depth of points in the scene. These estimates are represented in a stereo disparity map, which is constructed by matching corresponding points in the stereo pair.
Stereo vision is used in many applications such as robot navigation, 3D movie recording and production, object tracking, machine vision, and range sensing.
- Badges
- Science topic