AtTrack

Machine Learning to leverage the human learning process

Comment

Inspiration

With online lectures becoming a norm, it's important to know if students are really concentrating while learning remotely. It's next to impossible for professors to keep a watch on students while teaching them. A simple that provides insights on the average attention span of the class in order to help them adopt newer teaching methods that would ensure better understanding in online classes.

What it does

Understanding students’ attention span and what type of behaviors may indicate a lack of attention is fundamental for understanding and consequently improving the dynamics of a lecture. Various applications in analyzing students during online exams, analyze candidates during interview processes.

How we built it

The entire pipeline consists of the following key steps:

  1. Detecting a human in frame
  2. Detecting facial features and head pose of the person in frame
  3. keeping a track of any movements that take place in real time
  4. Gather user specific data points based on the unique ids allotted to each student
  5. Create visualizations to collect insights based on the data collected

Steps 1 and 2 required training ML models based on different open source datasets. An OpenCV filter was created to extract and track movements in real time in Step 4. Lastly, the generated data points were ported to AWS QuickSight for visualizations.

Challenges we ran into

Few of the problems faced during the entire length of hackathon were:

  1. Limited computer resources for training the ML models which consumed most of our hacking time resulting in compromising with the UI
  2. Ambiguity in face detection due to low light, face painting, jewelry, etc.

Accomplishments that we're proud of

A model that accurately detects students and tracks their movements in real time with nearly 84% accuracy.

What we learned

36 hours and a dedicated team alongside makes achieving any milestone a lot easier!

What's next for AtTrack

  1. Improve accuracy further by not misclassifying the environmental aspects such as low lighting, people or paintings in background and other aspects such as face paints, blue light glasses, jewelry, etc.
  2. Automatically record attendance for students joining class.
Share this project:

Updates