Project: Pizza Counting System

Project Definition

It all started with a simple question:

"Can i count how many pizzas are sold just by watching a video?"

At first glance, it seemed like an easy task. Point a camera at the kitchen, run a model, and start counting. But as i dove deeper, i realized this was more than just detection. It was about understanding context and covering a lot of edge cases. SO what really defines a pizza sale?

🧠 Asking the Right Question

As i explored different strategies, one critical decision had to be made early:

"What exactly does a pizza sale look like on video?"

First answer: We count a pizza is sold when the employee hand the pizza box to the customer.

At first, it made sense. That moment when a customer receives their box feels like the natural end point of a sale.

But there are ton of cases i have to address.

• What if the camera angles didn’t always capture that moment?
• The pizza could be ordered online
• Sometimes staff handed over multiple boxes at once creating confusion.

It was too fragile. Too dependent on things i couldn’t control.

Second answer: A pizza is considered sold once it is placed into an open box

• This was the breakthrough.
• I realized that the moment a pizza is placed into an open box is consistent, observable, and happens for every order walk-in, takeout, or delivery. It became the perfect "event trigger" for counting sales.

Now! Let’s get straight to the next phase:

We count a pizza is sold when employee put the pizza into box

🛠 Development Phase

• The heart of the system would be YOLOv11, the fast and powerful object detection model by Ultralytics.
• The input will be a video file (.mp4) or video streaming from webcam/cctv.