Inspiration

Due to the pandemic, we saw many businesses struggling with monitoring the customers entering the store and enforcing the current mask policy to reduce the spread of Covid. In parallel, over the break, we spent the time to learn AI and Neural Networks so we wanted to challenge ourselves by implementing out knowledge into a feasible project.

What it does

Uses the camera of the device, usually some type of security camera, to detect persons entering the premise and check whether or not they are wearing a mask, which is then sent to a security system to allow entry(demonstrated by an Arduino as motors for a gate).

How I built it

The core of the project is made using Python Django, Tensor-flow and Keras, built on a modified version LeNet-5 This was fed with a large pool if images of people wearing or not wearing a mask to train the network. This was along side the Arduino code to demonstrate an entry system and Front-End code for better UI.

Challenges I ran into

We used Alex Net and VGG-16 initially but they took more time to train and provided us with poor results. The communication from Py-Serial to Arduino was unnecessarily more complicated because of the protocols on the Arduino allowed only for a specific order of commands before it could be used how we wanted it.

Accomplishments that I'm proud of

Getting the neural network to function and yield accurate results. The overall cost of implementing our design onto current security systems would be low as all the business would need to do is to use our compiled software model and just connect the pieces to it, which would be an external camera and their security system.

What I learned

We learned how tedious and lengthy it was to train various large neural networks only to achieve subpar results (with the allocated time given). In addition, the creation and importing of CSV files were heavily resource consuming. This would be alleviated by using a custom python script to convert and store the files for us.

What's next for Mask-Off-Detection-System(MODS)

To improve the project, we would like to implement a notification system to alert the security team when there is an issue directly rather than needing to use the software. To further improve our neural network by recognizing people incorrectly wearing their masks so they can't fool their way into the premise. Modularity of the software so it can be used on all devices than just a computer.

Built With

Share this project:

Updates