The Problem
Driver drowsiness can be difficult to recognize before it becomes dangerous. For a university assignment, my teammate and I developed a desktop application that detects signs of drowsiness from facial movements.
My Contribution
I co-led the project with one teammate. We tested eye closure, yawning, and blink patterns as indicators of drowsiness. We also adapted a dlib facial landmark predictor to focus on the eyes and mouth instead of all 68 facial landmarks. Our goal was to make the model lighter and reduce inference time.
Challenges
This was our first computer vision project and our first experience conducting research. We had to learn the methods while deciding how to design the experiments and build the application.
Results and Reflection
We completed the research assignment and built a desktop application. Looking back, I can identify limitations in how we conducted the research. Being able to recognize those limitations now is also a measure of how much I have learned since the project.