Agisoft Photoscan

If you have been following this blog for some time, you will know that when it comes to Photogrammetric reconstructions I have always been a strong supporter of 123D Catch by Autodesk. I find that it is by far the easiest program to use, yet the results are still amazing.

Recently they have released a paid Pro version that provides all the same results but allows the program to be used commercially. I think that is utterly brilliant, as it means it doesn’t halt research, but it still allows revenue for the developers if the user is himself making money from the software.

Having said all this, I’ve started investigating new software, to see if there is anything that can bring improvement to what I already have with 123D Catch. Up to now, the best solution I have found is Agisoft Photoscan, which I had already used in the past but not to its full extent.

Image

Previously, I only managed to create low quality models, due to an error in the program, but I have now managed to make really good models of both objects and features. As such, here are the pros and cons of Photoscan compared to 123D Catch:

PROS

  • Quality: This is the best starting point. With medium and high settings the quality can be really good, and generally you can get many more points than 123D Catch. With lower settings however the results do fluctuate.
  • Control over process: If you are looking into more complex Photogrammetry this passage is very important. In 123D Catch you upload the images and get the result. That is it. In Photoscan you can go through all the stages (photo alignment, sparse cloud, dense cloud, mesh, texture) and change the settings to improve the finished product. You can even export the points as they are, and alter them using other software. It allows much more flexibility.
  • More tools: You can select points, create masks, remove points based on certain parameters and more. Often these are not needed, but on occasion they can be just what you require.
  • Many photos: 123D Catch struggles when you upload more than 50 photographs, and the results suffer. In Photoscan this issue doesn’t exist, and you can easily make models with hundreds of images. This is perfect for making large scale models, as the more images you have, the more accurate it is.
  • No internet required: Often you find yourself in situations in which the internet is not great or is totally non-existent. Photoscan doesn’t need a connecetion to an external server to process the model, so even if the computer is not connected, it still works.

Cons

  • Paid software: Although as software goes this is well priced, for people doing non-commercial work it is always difficult to have to keep up with the expenses for programs.
  • Memory: 123D Catch uses an external server, which means it does not use your own computing power. If you are trying to get high quality models, Photoscan will put a serious strain on your computer, and often it will even crash due to insufficient memory.
  • Time: I can get a model done in 123D Catch in 5 minutes if the internet is good. The worst I have ever had is probably an hour. With Photoscan I have had to wait actual days for models to be complete, and sometimes I waited many hours before getting an insufficient memory message.
  • Simplicity: If you are just starting out with Photogrammetry 123D Catch is still the easiest of all programs I have used.

Overall, I think I would use 123D Catch for small scale and quick stuff, while Photoscan would be useful for large scale, or when I am trying to research something specific, so I am more in control of the results.

Image

Advertisements

Program Overview: 123D Catch

Image

123D Catch is my program of choice when it comes to Photogrammetry, and although it comes with some limitations, I still believe it is the most efficient piece of software when it comes to everyday archaeological modelling. It combines speed, accuracy, an easy interface and it is free to use, even commercially.

Speed: Especially in commercial archaeology speed is everything. You have a job to finish and a limited amount of time to do it, so the aim is to cut down the time spent recording while still preserving the same amount of information. 123D Catch is extremely quick provided it has a good internet connection. A basic model with 20 phoitos can be processed in ten minutes, and a more complex one in 15. Progrograms like Agisoft Photoscan take much more time, around 45 minutes for simple models, and many hours for larger ones. In addition to this 123D Catch models are processed o nan outside server, meaning you can run many diffferent models at the same time without slowing down your computer, while Photoscan really puts a strain on it. Othe rmethods of Photogrammetry, like those described by Ducke, Reeves and Score (2010), take days to process entirely. Similarly, a good laser scan can take a long time to process, and many stations may have to be set up.

Accuracy:  I’ve talked aout his before, so I won’t discuss it much here. Compared to other programs (and laser scanning), 123D Catch does seem to trail a bit behind when it comes to accuracy. The error margin seems to be between 1% and 0.17% (see table in earlier post and Chandler and Fryer 2011). Stereo closed range Photogrammetry has a range of 0.1% and 0.01% (Chandler and Fryer 2011) and laser scanning has a general standard deviation of 0.05-0.02% (Boehler & Marbs 2004; English Heritage 2011; Kersten et al. 2005). The main question to ask though is: do we need an error margin better than 1% for what we are using it for? Up to now I have had no problem with 123D Catch’s accuracy, so I feel that the answer to that question is no.

Easy interface: I’ve always felt strongly that this type of technology should not simply be used by professionals who spend ages learning how a program works, but that it should be accessible to everyone. Therefore it is important the software is simple to use, with a good interface. With 123D Catch all it takes to make a simple model is click the upload button and then the process button. When it comes to more complex features of course it’s not as simple, but for everyday use it is ideal. The Ducke, Reeves and Scores method is extremely complex, and requires to user to know how to code and to use four different programs. The Photoscan interface is simpler, but the navigation tools still could do with improvement.

Image

Freeware: Not much to say about this. It is free, so accessible. It provides the same results as expensive programs and anyone can own it.

Limitations: Of course there are also some limitations to the program. The main one I have found is that unlike Photoscan, it relies on an internet connection, which makes it difficult to use on a site. While digging on Ham Hill I had to pop down to the local pub every night and use their extremely slow internet connection to process models, which would take ages. 123D Catch also struggles with larger features, mainly due to the fact it doesn’t work well with more than 60 or so photographs. There are ways round it, like manual stitching, but it always requires more effort than should be necessary. Manual stitching itself is temperamental, and half of the times it takes a long time to find points in an image to then get an error message.

Overall I would still recommend it as the best program for Photogrammetry.

References:

Boehler, W. and Marbs, A. (2004). Investigating Laser Scanner Accuracy. Available: http://archive.cyark.orarchive.cyark.org/temp/i3mainzresults300305.pdf. Last accessed 9th May 2013.

Chandler, J. and Fryer, J. (2011). Accuracy of AutoDesk 123D Catch? Available: http://homepages.lboro.ac.uk/~cvjhc/otherfiles/accuracy%20of%20123dcatch.htm. Last accessed 9th May 2013.

Ducke, B., Score, D. and Reeves, J. (2010) Multiview 3D Reconstruction of the Archaeological Site at Weymouth from Image Series. Computers & Graphics, 35 . pp. 375-382.

English Heritage. (2011). 3D Laser Scanning for Heritage . Available: http://www.english-heritage.org.uk/publications/3d-laser-scanning-heritage2/3D_Laser_Scanning_final_low-res.pdf. Last accessed 9th May 2013.

Kersten, T., Sternberg, H. and Mechelke, K. (2005). Investigations into the Accuracy Behaviour of the Terrestrial Laser Scanning System Mensi GS100. Optical 3-D Measurement Techniques VII. 1 (1), p122-131.

Image