In 2006, my friend Myoung Lah and a group of housemates set a tripod atop Baker House at MIT and painstakingly moved the camera by hand at regular intervals to produce about 1,000 different photos from 6 distinct angles along the Charles River. They apparently did this in shifts for almost 24 hours. I'm still not really sure why they did it or why they didn't automate it to ensure consistency in the shots. By this time, though, I had graduated and moved back to Atlanta so I did not have the opportunity to ask Myoung: “why did you do this and why didn't you automate it to ensure consistency in the shots?” It wasn't until Myoung reached out to me asking for some advice on a web app he was building to display the images that I even knew they existed. I happily helped him with the problem he was having and, in exchange, I asked if he would send me copies of the photos. A few weeks later, a DVD arrived in the mail and I discovered some truly incredible shots of the Charles River. I remember wanting to do something with the photos back then, but I stuffed the DVD into a drawer and promptly forgot it existed.
But last month, almost 8 years later, I began preparing for a move to North Carolina and found the forgotten DVD tucked away in a stack of other discs. I immediately put it in my DVD drive to make sure it was still in good shape and promptly copied the photos onto a shared drive so I could easily look through them. I had failed to do anything constructive with the images the first time around, so I figured I would try to assemble them into something remotely interesting before I had a chance to forget them again. There was a particular sequence from about 5:30am until after 8am when the sun can be seen rising over Back Bay. In one these photos, you see the sun make its first appearance by sneaking in between two buildings in the skyline. The only issue was that despite their best attempts and despite having a tripod, the camera angle varied quite a bit even over the few hours of the sunrise sequence. I wanted to take a slice from each photo and build a composite image with the time progressing from left to right so that on the left you could see the darkness of pre-dawn progressing through the bright oranges of sunrise and finally into the blue skies of daytime. I hoped with enough tweaking, I could even feature the sun peaking through the buildings in one of the slices. But first I needed to deal with the photo angle issue. I optimistically tried some naive approaches, then considered just manually aligning the images, but finally I took an algorithmic approach. I will save the gory details of that for another post, though. I will note, however, that my goal here was not produce a seamlessly stitched image. In fact, I try to avoid any transforms that would distort the original image. I was thinking of this more as a way to organize the original photo sequence than a way to magically smoosh them into one. To define this more technically, my image registration is based on just a translation matrix and not a general homography matrix.
So what you see above is a sequence of images taken over a three hour window on the morning of November 3, 2006. Clicking on any of the slices will show you the original photo still aligned with the other images. If you like, you can also move between original images. In my time at MIT, I probably saw far too many of these sunrises. This would be about the time I left the lab many mornings. Thanks to the effects of sleeplessness on memory, though, I remember fewer of them. It's nice to have one so vividly preserved. And thanks to Myoung and friends for moving that camera (even if we still have no explanation for why they did it).
Thanks to Stumm for reviewing early versions.