6-7pm Queue

• 

  Long queue with some people queuing for other stalls

• 

  Many people but none queuing for the western stall

• 

  One queuing for the western stall with some nearby

Inspiration

At dinner time, around 6-7pm, our residential college's dining hall gets really crowded, especially when good meals are served that day. Me and my friends often find myself walking all the way to the dining hall only to find ourselves stuck in the very long queue so we thought wouldn't it be great if we could see the queue at the dining hall and choose whether we should go now or wait it out?

What it does

6-7pm Queue helps us see the real-time crowdedness of the dining hall with computer vision by analysing the current queue for each stalls. For the hackathon, it's still in the proof of concept phase so we focused on getting the computer vision to be tweaked to analyse one particular stall of the dining hall, with someone being required to take a picture from our pre-determined angle that captures the queue for that stall.

How we built it

We built it with Next.js and Python, leveraging on Ultralytics Yolo26 real-time object detector model. We also used Supabase for our simple database management to store information analysed from each capture (still image) of the queue.

Challenges we ran into

Finding the right metrics for determining whether or not one person is in the queue was not easy as from the sample pictures of the queues that we took, it can be quite messy with people having their faces not visible, people not standing in one straight line in the queue, and people standing in other queues within the frame of the target stall's queue.

Accomplishments that we're proud of

We were proud that we could utilise the model to detect multiple human body parts and use that to determine a rather accurate conclusion on whether or not said person is in the queue, including facial direction, ankle direction, relative distance from other people in the queue, and more.

What we learned

We learnt that it is very important to try as much as possible to tidy up our code as we build our product as some functions can go very long when not careful and when there were loose magic variables, it could get quite confusing.

What's next for 6-7pm Queue

• Fixed camera for a fixed angle that can monitor multiple queues of the dining hall
• Add the necessary measures to allow all queues within the dining hall can be monitored
• Add automatic image upload (from the fixed camera) so that the dining hall staff need not take pictures every now and then to update the queue status

Built With

• next.js
• python
• supabase
• yolo