Computer Vision Research Papers - Academia.edu (original) (raw)

Computer Vision Photogrammetry allows users to simply upload a series of overlapping pictures of a scene into dedicated software in order to automatically generate an accurate three-dimensional model of that scene. In this master thesis I... more

In the introductory chapter I first discuss the importance of archaeological site recording, as well as the importance of exploring new and perhaps better ways of documenting our heritage. I then stress that for a new recording method to see widespread use within the field of maritime archaeology, it must offer distinct advantages over traditional manual recording methods, but it must also be accessible, both in terms of costs and in terms of the technical expertise required to use the method. I believe Computer Vision Photogrammetry could meet these requirements, and as such I think the method is worth investigating further.

In the second chapter I then provide readers with the necessary background information in order to understand what photogrammetry actually is, how it works, and how it has evolved from the 19th century to the present day. Special focus is put on the new generation of Computer Vision Photogrammetry techniques, and on explaining what sets these techniques apart from photogrammetry approaches used in the past.

Next, in the third chapter I have a closer look at past uses of photogrammetry in underwater archaeology. Photogrammetry has been used for underwater archaeological site recording for several decades now, in some cases as early as the 1960s. Nevertheless, the method has never broken through as a mainstream recording method within our discipline. My literature review demonstrates that while photogrammetry had significant advantages over traditional manual recording methods, in the past the method was too technical, too expensive and too unreliable for most archaeological projects.

For the remainder of my dissertation the challenge is therefore to assess whether modern Computer Vision Photogrammetry can successfully overcome these past obstacles. In order to draw some well-founded conclusions, I want to use Computer Vision Photogrammetry software to process a number of archaeological case studies. As a first step, in the fourth chapter I therefore attempt to make an informed decision on which specific photogrammetry software to use to process these case studies. After testing various software packages, and after assessing the relevant scientific literature, my software of choice is Agisoft PhotoScan.

In the next chapter PhotoScan is then used to process data from three archaeological shipwreck sites. The intention is to really test the reliability and flexibility of modern photogrammetry and as such, rather than using ‘perfect’ picture sequences, my case studies focus on the use of particularly challenging or unconventional input data. In the first case study a shipwreck is modelled from ‘casual’ video footage that was not originally captured for photogrammetry purposes. In the second case study a three-dimensional model is created using ‘legacy data’; archived pictures originally taken more than 20 years ago. In the final case study a shipwreck is modelled using images captured in a low-visibility environment. In each case study the site is first briefly introduced, the main challenges of the specific dataset are then outlined, next the processing procedure is discussed and finally the results are presented.

In the sixth chapter I then discuss the results of these case studies as a whole, and attempt to draw some conclusions regarding the overall impact of new developments in photogrammetry on our discipline over time.

I conclude that compared to traditional manual recording methods, Computer Vision Photogrammetry is capable of significantly reducing underwater recording times while simultaneously producing more accurate, more detailed and more objective three-dimensional results. Furthermore, compared to past photogrammetry approaches, today Computer Vision Photogrammetry is more affordable, easier to use, more time-efficient and more reliable. Thanks to the confluence of a) reduced costs and b) increased performance, I believe a ‘tipping point’ has finally been reached. For most underwater archaeological projects the cost of using photogrammetry will now actually be lower than the cost of using manual recording methods. Considering the fact that photogrammetry also produces better results than manual recording methods, we have every reason to believe that Computer Vision Photogrammetry will play an increasingly prominent role in the field of underwater archaeology in the years to come.