A #SmartCityHack project rolled by Alex, Daniel, Steven, and Zack!
Abstract: using a live feed provided by the City of Auburn, we can programmatically determine if Auburn Tigers fans have begun rolling Toomer's Corner. This program tweets at @IsToomersRolled, which is used by our web app at https://istoomerscornerbeingrolled.herokuapp.com/ to display either "No..." or "Yes!".
The core of our project is theAppThatLetsUsProcessToomersCornerBeingRolled, which uses the Haskell functional programming language to watch the Toomer's Corner webcam and assess whether it is currently being rolled.
We used the friday and friday-devil Haskell packages to do our image processing. Both of these depend on the C library DevIL (Developer's Image Library).
Our script first grabs stills from the live feed which are fed into our Haskell code. We use the Canny edge detector to get a stripped-down image with just the sharp edges. (WARNING: a mathematician approaches!) We think of the edge-detected image as a (discretized) real-valued function of two real variables. The Sobel operator on this edge-detected image gives use a (discretized) gradient vector field. We then compute the direction of each gradient vector and count up the number of more or less horizontal (within about 10 degrees of true horizontal) gradients. The number of horizontal gradients is the number we use to determine whether Toomer's Corner has been rolled. Our reasoning for using this number is that strips of TP hanging down from trees and lamp posts (and everything really) will give us a large number of horizontal gradients because there are a large number of vertical edges. At least this seems to be the case - our testing generated no false positives. However, we'll need to wait until the next big ballgame for more rigorous field testing.
Once we've determined that Toomer's is being rolled, we've got to get the word out. (After all, how ELSE could people know?) That's where theTinyAppThatTweetsAPhotoWhenToomersIsRolled comes in.
Rather than rolling our own server to persist the good news, we cut to the chase by using Twitter. The Twitter API has several restrictions (which we'll describe in more detail in the next section); however, since our detection app will request a tweet very infrequently, we won't hit any issues here.
Using the Twitter RubyGem library, we may easily authenticate with the Twitter API 1.1, and post a message to @IsToomersRolled affirming the rolling, as well as the video still which triggered our program.
We decided we should also create aWebAppThatCanCheckTwitterForYouToSeeIfToomersIsBeingRolled in the vein of similar projects such as https://isitchristmas.com/. It's hosted by Heroku at https://istoomerscornerbeingrolled.herokuapp.com/.
Since we're using Twitter as our persistence layer, we need to interface with the Twitter API again to detect what our latest tweet was. If we can do that, the rest is easy: if the last tweet was within 12 hours, we assume the rolling is still relevant, so we display the tweeted (twatted?) message. Otherwise, we just display "No..." and move on.
PROBLEM! The new Twitter API (1.1) is very stringent about how many requests it will accept from an application before timing out: about one request per minute! It also requires authentication, even though all our tweets are public. Therefore our original plan to develop a Single Page App to read the feed with pure Javascript didn't fly.
We returned to Ruby to solve this problem, opting for the lightweight Sinatra platform rather than a more heavyweight solution like Rails. Using the Twitter RubyGem again, we can authenticate and pull our latest tweet on the backend, and use a simple ERB template to display the appropriate response.
The hardest puzzle here was how to circumvent Twitter's rate limit, without
adopting a heavyweight/redundant database in our application.
Enter the flat-file persentence pattern: on each request, we check to see
if .twitter_cache was written within the last 75 seconds. If not,
then we make a new request to the Twitter API, which we write to our
cache file for use within the last 75 seconds. ActiveRecord, eat your heart out.
Since the Twitter API sucks, we provided our own API for the
cornucopia of applications which could use this valuable data:
https://istoomerscornerbeingrolled.herokuapp.com/api/. Just remember
to check that data["stale"] == false before displaying an outdated
message!
Following our work at the regional competition, we've developed a prototype to analyze footage of Toomer's Corner using Principal Component Analysis rather than counting horizontal gradient vectors. Information on that project may be found at its README page.
All our projects within this organization are MIT licensed. Go crazy, kids.