In the last post, I discussed the capturing phase for the first case study of the NUIM-AFF621: Remaking the Physical – 3D recording assignment. This post will examine the processing of the captured images with Agisoft Photoscan Professional. The captured images are of a bipartite bowl (2004E505:5) found in a cist grave in Liscooly, Co. Donegal. The images are courtesy of the National Museum of Ireland (Archaeology) who supported the project and provided access to the artefact. A detailed description of the vessel is presented in Post 2.
I downloaded Agisoft PhotoScan Professional software for a 30 day free trial. “Agisoft PhotoScan is a stand-alone software product that performs photogrammetric processing of digital images and generates 3D spatial data” (Agisoft Website). It is suitable for close-range photogrammetry and creating 3D models of cultural aretafcts. It depends on a large amount of overlapping photographs to “serve as an input for the program”, these are “aligned and dense point clouds or DSM models are generated” (Svatý 121). Agisoft recommend the following system requirements in order for the product to work: Windows XP or later (32 or 64 bit), Mac OS X Snow Leopard or later, Debian / Ubuntu (64 bit) with an Intel Core 2 Duo processor or equivalent and 2GB of RAM (Agisoft PhotoScan User Manual). Although, in order to process 200-300 photographs, they recommend 12GB Ram.
I was working off a HP 250 lap top which satisfied the minimal recommendations at least. Agisoft Photoscan Professional was easy to download and install, and was activated with a licence key that was sent to me by Agisoft via email. The tutorials from Samantha T. Porter are a good starting point for a photogrammetry neophyte, as well as the Agisoft PhotoScan Professional guide for beginners “3D Model Reconstruction.” From these sources, I organised a general workflow for processing photographs in Agisoft PhotoScan Professional to create a 3D model as follows:
- Check Preferences;
- Import photos;
- Mask photos;
- Align photos;
- Build dense point cloud;
- Build mesh;
- Export results.
I created an Agisoft folder with a copy of the 138 photographs of the Liscooly bipartite bowl (2004E505:5) to use for processing to create a 3D model. Using the Agisoft beginners tutorial as a guide, I set the recommended preferences, and imported the photographs. I checked the images for their quality, and was satisfied that they met the requirements being in the region of 0.59-0.7. Then I began the daunting task of masking the images. While the magic wand tool was useful, I still needed to use the rectangle and scissor tool to mask some elements from the scene. With the images masked I clicked to align the photos in the workflow menu.
In the camera alignment stage, PhotoScan searches for common points between the photographs in order to match them and also assess the position of the camera for each photograph to create a set of camera positions. In the meantime, while Agisoft PhotoScan was running, my laptop was in over drive with notices for high disk/memory usage and I had trouble working with other applications, even Word documents were freezing. Alignment took about six hours to complete. When it finished I noticed a large number of the photographs did not align, particularly those of the inside of the bowl. I still went ahead and followed the guideline in the beginners tutorial to build the dense point cloud and the mesh for practice, and ended up with half a bowl in the first attempt.
For the next attempt I opened the first project dataset again and re-saved it as Version 2. Referring to the Agisoft PhotoScan User Manual (Professional Edition, Version 1.0.0), I read through the reasons why photographs do not align and I deleted some of the photographs from the chunk in the programme if the subject did not fill the frame sufficiently. I reset the camera alignment, and I began the alignment process for a second time with 126 masked photographs. Yet again, some of the photographs did not align and the model is shown below.
As some of the photographs were quite dark, as a result of bad settings during the capturing stage (discussed in the previous post), I set out to alter the brightness and contrast of the photographs and attempted to create some uniformity in texture. I opened the application and saved a new project as Version 3, and uploaded the photographs that had been rectified. I checked the image quality and commenced masking again as they were a new set of photographs having been altered for brightness and contrast. Once again, not all the photographs were aligned, and the representation of the vessel was still very unsatisfactory as shown below
At this point I was frustrated and conscious that time was scarce, so, I decided to try out 123D Catch to create a 3D model from the captured data. After working with 123D Catch, which I discuss in the next post, I returned to Agisoft PhotoScan for another attempt.
Further reading of the Agisoft PhotoScan User Manual revealed the possibility to place markers on photos that did not align automatically, although, I found the description of the process in the manual somewhat vague. Nonetheless, Samantha Porters tutorial (Part 2) was beneficial here.
The purpose of placing markers is to assist in the alignment of photographs that do not align automatically via the software. However, this was a very slow and formidable task. First, it was difficult to place markers due to the repetition of the vessels design. Second, this had to be done viewing one photograph at a time. Unlike 123D Catch where it allows for three photographs to be viewed simultaneously in manual stitching, and thus, provides a better chance of placing manual points in the same place in all three photographs; however, this is not possible in Agisoft PhotoScan. Rather, the user has to keep moving to and from the photographs that need to be aligned in a single image view, and then match these markers with a photograph that has aligned automatically. I then accidentally closed the programme without saving and lost the markers I had placed, and so I had to start again.
In my final attempt (Version 5), I placed the markers as best I could and with time against me I was eager to run the alignment with the markers in place to see if there was any improvement. Indeed, there was some enhancement in comparison to the previous attempts, also I learned from the Samantha Porter tutorials – how to get rid of loose dense points with the clean up tools, and this was applied before I sent the model for a final mesh. However, as a photo-realistic representation of an ancient artefact, the final 3D model is well below par. Indeed, I would not consider the final model worthy for 3D printing as I would not insult the fine craftsmanship of the potter who made it 4,000 years ago.
Finally, Agisoft PhotoScan Professional supports the direct uploading of a model to Sketchfab and Verold for display as a 3D model online via the file menu. However, when I tried to upload the model to Verold, PhotoScan notified me to say that the API key was invalid, even though I had received the API key when I registered with Verold. Thus, I uploaded the obj. file to the asset library in Verold and the online model is available here.
On reflection, had I used place markers in the first attempt with the original 138 photographs – would it have made any difference? I did not have the time to check that out, although I will most likely try it out for my own satisfaction and to get better acquainted with the software. Though, I am inclined to think that the problems were related to the capturing of the object, i.e. the photographs, and so, this was not the fault of the software. Thus, I will conclude that careful attention needs to be given to the capturing phase of a digital recording in order to produce “Good” photographs as I propose that this is fundamental to creating a successful 3D model with Agisoft PhotoScan Professional.
- Agisoft PhotoScan User Manual Professional Edition, Version 1.1, 2014 www.agisoft.com. Web. 10 Mar. 2015 <http://www.agisoft.com/pdf/photoscan-pro_1_1_en.pdf >
- Agisoft. ‘Tutorial (Beginner level): 3D Model Reconstruction with Agisoft PhotoScan 1.1.’ www.agisoft.com. Web. 10 Mar. 2015 <http://www.agisoft.com/pdf/PS_1.1%20-Tutorial%20(BL)%20-%203D-model.pdf>
- Sváty, Z. “Use of Photogrammetry in Forensic Science.” XIIIth Youth Symposium on Experimental Solid Mechanics, June 29th – July 2nd, Czech Republic. Web. 26 Feb. 2015. <http://www.itam.cas.cz/ysesm2014/proceedings/id_38_ysesm2014_proceedings.pdf>