Let’s get snapping!
We all know that once you’ve posted something online you can never get it back, and that somewhere on the internet your photos live forever. You might cringe at the thought of those embarrassing pictures of you from birth through umpteen children’s parties to even more embarrassing adult parties, and regret the day you published those 158 photos of your holiday trip down the Thames to Greenwich, but what will historians of the future make of the millions of photographs that are posted on the internet and how will they analyse them? Time-lapse mining might be one way.
In time-lapse photography your camera is fixed in one place, you take a picture…wait, take a picture…wait, take a picture…etc. Once you have created a reasonable amount of footage you then compress and compile it into a bite-sized mega-speeding version of what you have photographed. Time-lapse photography is fun for the occasional day – perhaps you set it up to monitor the Geminids meteor shower the other night and created a great star-trail photograph – and it can be useful for automated photographs via CCTV; but how can it be used to help analyse the content of the internet?
Data mining is the computerised process of sorting through large sets of data, identifying patterns, establishing relationships and using them to solve problems through data analysis. This is usually used as a tool to predict future trends. It has been used recently to create ad campaigns for companies like Netflix and Spotify, based on their user data – but many people feel very “Big Brothered” by the idea of companies mining their data. https://www.nytimes.com/2017/12/17/business/media/netflix-spotify-marketing.html
Ricardo Martin, David Gallup and Steven Seitz, however, are developing a technique which creates time-lapse sequences from photos that people post on the internet – in a way a tool to look at the past instead. It sounds simple – find photographs and stitch them together – but actually involves sifting through tens of millions of photographs, sorted by timestamp and geolocation; an enormous computational task of hundreds of terabytes. The photos are then lined up to look like they were shot from a single vantage point – which involves a major amount of digital trickery, or from a virtual camera moving continuously in time and space to offer a 3D parallax effect.
11,000 time-lapse sequences have already been created – some, like the erosion of Briksdalsbreen Glacier in Norway and the rise of New York City’s Goldman Sachs Tower work well, others rather lack the wow factor. However, this is the dawn of a technique that will be a valuable tool for anyone who wants to study the effects of time on different environments, without having to leave cameras in place for years.
Their first video, produced in 2015, contains 86 Million tourists’ photographs. Read more about the technique and challenges at http://grail.cs.washington.edu/projects/timelapse3d/ and view Time-lapse Mining from Internet Photos SIGGRAPH 2015 by Ricardo Martin Brualla in our header video (note the audio is very quiet) or look at their “3D” video below and see how they are developing their technique with parallax.