Better Studies Through Facial Recognition & Machine Learning
When Dylan Greenberg ’21 chose “Project Insight” as the name of his senior computer science project it was a tongue-in-cheek reference to a Marvel Cinematic Universe plotline that involves nefariously tracking humans around the globe by satellite. While his project does involve potentially alarming phrases like “facial recognition,” “machine learning,” and “artificial neural networks,” there’s nothing so insidious at play here. On the contrary, Greenberg’s goal was to build a tool that would help students stay organized and could conveniently access their academic information all over campus.
Project Insight is a student information kiosk that is linked to the Student Information System (SIS) and Gould’s Learning Management System (LMS) Canvas. Utilizing both of these systems Greenberg created custom software in Python and NodeJS that pulls schedule and assignment data and displays it on a screen at the kiosk. The touchless station doesn’t require any input from the user though. Other than their face. The camera on the kiosk is very good at determining the presence of a face. Knowing who that face belongs to is much more difficult.
“I had to make my own facial recognition model…which was a journey,” says Greenberg. “This was my first real deep dive into machine learning. I had to make a server that each station could offload the heavy processing onto so that the kiosks could be more compact and easier to set up.”
That is to say, the camera doesn’t ID the student. It takes an image of their face, crops it, and sends it off to the NVIDIA Jetson Nano that has a neural network running on it that can recognize the face. An important nuance is that the Jetson isn’t just referencing a gallery of pre-loaded student headshots. The neural network has already learned what each student looks like from their directory photos and actually recognizes them, and returns their personal info to the screen. That’s machine learning. It’s worth noting that the Jetson was initially recommended to Gould as a tool for computer science projects by John Drew ’14 who is currently an engineer at NVIDIA.
Computer Science Teacher and Gould’s Director of Technology Jason Chase says Greenberg’s abilities are impressive.
“He’s able to quickly translate an idea into working software. Developing the kiosk alone was a great feat for a student at the high school level. Adding a multi-layered distributed artificial intelligence component was far beyond my expectations.”
For now, two kiosks exist. One in the McLaughlin Science Center, and the other in Sanborn Family Library. But with the programming completed, setting up more stations around campus would be relatively simple. Future upgrades could integrate REACH, the management system that allows students to check into dorms and academics buildings.
Could Project Insight eventually live up to its namesake and start tracking your whereabouts from space?
“It’s always been in the back of my mind that it’s possible,” says Greenberg smiling, “however, it’s not practical [for now] to have facial recognition running on that large a scale. It’s really just useful in niche scenarios.”