It was lonely being the only ones to see the among us crewmate everywhere we looked. That's when we decided to spread our gift with the world by creating an app that utilizes computer vision to detect and draw Amogi (the plural of Amogus, of course) in an image.

What it does

Amongi are everywhere. We've created an app to help see the unseen. Those AMONG US who hide in the darkness, who vent. Come, open your third eye, with Amogitech.

How we built it

  • Custom in-house JavaScript/HTML5 Canvas tools for "annotating" our data set
  • Developed a proprietary binary format for storing various drawing animations (our .sus files)
  • Created an engine for replaying drawings, utilizing expanded transformation matrices to best fit any image
  • Used Python's OpenCV library to heuristically analyze different images, and infer where Amogi could be hidden

Challenges we ran into

The image detection process was easily the most challenging aspect of this project. A preliminary approach was to use a Sobel convolution filter in combination with Circle Hough Transform to detect circles within an image. There were, however, too many false positives in noise-ridden images, so we spent a good half of our hacking period honing and refining a better heuristic for increasing the contrast in photos and performing contour detection with ellipses bounding volumes.

Accomplishments that we're proud of

We were able to design the logo to imitate Logitech's. This took some time out of the coding process, but the end product speaks for itself. Additionally, amogus.

What we learned

This was our first time developing tools for computer vision. We learned a lot about cleaning and transforming our image datasets into useable formats for our algorithms. We also tend to neglect our graphical & user interfaces, so this time we spent extra time learning how to use CSS animations and basic design principles.

What's next for Amogitech

The next reasonable step for Amogitech would be the introduction of Amogitech 2! AMONG the features will be live video Amogus detection, as well as a confidence interval that better identifies Amongi.

Share this project: