As Ontario businesses begin to slowly reopen during the lockdown, many can't help but feel unsafe due to the current global pandemic. To slow down the spread of the viruses, businesses must enforce mask rules, and ensure that those with a fever (38 C or higher) are properly self-isolating. Ensuring that people wear masks requires an employee posted at the entrance to slowly let people in one at a time. Also, many stores neglect measuring temperature, and instead will rely on self-reports from customers.
What it does
Tami (short for Temperature And Mask Imager) is an intelligent solution meant to be posted at the entrance of a physical storefront. As customers approach Tami, she is capable of detecting face masks, and checking body temperature from a distance! Armed with this information, she is capable of automatically rejecting people from entering the store. No humans required!
How we built it
Tami is built around the Raspberry Pi 4 (8GB). For thermal detection, she uses the AMG8833 8x8 thermal array, and a 5MP PiCamera 2 for face mask detection. A 7" touchscreen is used to show the live camera feed and provide visual feedback to the customer.
Tami speaks in Python, and OpenCV and TensorFlowLite is used for the machine learning/face mask detection.
Tami's outer enclosure was designed with Fusion360 and 3D printed in PLA .
Challenges we ran into
Time! Especially since 3D printing is a very time consuming process, we only had one shot for producing the enclosure (luckily for us it worked!).
Also Tami struggles with the machine learning model even with an overclocked Raspberry Pi. The performance is just barely good enough, but it could definitely be better! We had wanted to use hardware meant specifically for machine learning purposes (Coral Dev board), but they are notoriously difficult to purchase lately.
Accomplishments that we're proud of
We're proud to have actually finished something (something that works too!), this is our first ever Hackathon!
We're proud to have gotten a chance to participate in something with so many talented people, and we're extremely grateful to the organizers of the event for even letting us participate in the first place!
We're also proud to have built something that tackles a real life problem that is extremely relevant to us today!
What we learned
We learned (through trial by fire) how to manage tasks in such a short time frame, and how to prioritize an MVP (minimum viable product) before implementing the bells and whistles.
This was also our first "dip" into the metaphorical "machine learning pool" and we're definitely anticipating our next foray into the deep end!
What's next for Tami: the Intelligent Mask/Fever Screener!
We definitely hope to bring new features and improve upon Tami in the near future. Something already being planned is implementing Tami as an MQTT client so that it would be "IoT capable" and can communicate with a whole suite of other sensors.
As mentioned before, we are also trying to get our hands on Google Coral products, to dramatically increase the machine learning performance of Tami!