In this video, I will build a computer vision prototype in less than 48 hours. I will show you the processes that will easily and easily lead you to complete your project in a short time.

You will understand that it is no longer impossible to realize your ideas even just on the weekend or to deliver on time the great and awaited project for your company that you have been thinking about for years but you never built.

Developing a prototype is used to put into practice a proof of concept very quickly to understand if it works and how it works, only after we can go deep into the project and create the final product.

Problem to solve with Computer vision prototype

The first step in developing a computer vision prototype is to define the problem to be solved. In this video, I have taken as an example to follow a balanced diet according to the indications of a nutritionist.

To check the intake of calories for each meal I used a simple app from the smartphone, and I can check the calories of each food on the plate but it has the problem of having to manually enter each individual food and having to manually calculate the weight.

What if you could calculate the calories of a dish simply by framing it with the webcam?

Idea development and solution to the problem

To solve the problem I thought of a computer vision software that connected to a scale and a camera can tell me what is on the plate and the total calories. Below is a small outline of how it could be developed.

Computer vision prototype scheme

Having identified the general concept, I start thinking about the software and hardware side, always keeping in mind that this is a prototype for the moment and must work as the first goal.

Software for Computer vision prototype

Based on my experience I choose to use artificial intelligence with one of the most common frameworks to be able to correctly classify objects. I then take some photos and proceed with the training.

dataset

Fortunately, I save a lot of time thanks to ready-made development environments that help me a lot in this process.

yolo v4 notebook

I do a quick test to see the reliability of my model and as in the photos, it looks pretty good. However, I can further improve the model simply from the notebook

yolo v4 test

Hardware for Computer vision prototype

To proceed quickly and super efficiently I had thought of using Nvidia Jetson Xavier for software processing but in the end, I opted for simple raspberry pi. Given the limited time available for this test, I had to choose the readiest hardware solution at the moment.

I then connected the raspberry pi to a scale, a webcam, and a small touch screen

hardware computer vision prototype

Got the hardware configuration, gauge the scale and I’m ready to proceed to the next step with the Computer vision prototype

calibrate

Overall test

I put all the elements together and proceed with the test. As you can see in the image, I get the weight of the kiwi, which is automatically identified, and the calories.

Final remarks

As you can see, building a working computer vision prototype in just 48 hours is not an easy process and if you don’t have the right tools at your disposal, you will most likely be doomed to fail.