A Better Body Temperature Screening
#WFH suddenly becomes a real thing; many companies and officials compromise to do one or two of the working cycle remotely. This You-Know-What pandemic will change at least five to ten years of our way in interacting with people.
So, maybe you wonder how exactly we will continue to physically present in our office after the line of infection is flattening. Will the new standard protocols could force a new habit of hygienic lifestyle? It seems to be that the most crowded place manager has to think creatively to protect its community when working from home is no longer a must.
A Finger Connected to Fingers
Take the fingerprint attendance system as an example. We put the finger on it after our friends put fingers on it, this seems not promoting a hygienic lifestyle, but eliminating it can make a critical problem for a company that relies on attendance to calculate payrolls.
Let us jump to the body temperature screening procedure. Typically, any companies will assign few guards, and each of them carrying a thermo-gun, stopping all employees and visitors, aim it to their forehead, and when the result shows over 37.5°C, today might not be your best day.
Those two examples above made us think, what if we can provide a face recognition-based attendance system with body temperature screening integrated on it. It has to be a standalone and self-operated device, and by eliminating its operator, the solution will allow all companies politely to screen all persons without jeopardizing an employee’s health.
David’s Tools to Solve Goliath’s Problem
At Enygma, everybody embraced to think simple, and the first sensor for this solution that across our mind is the tiny thermal sensor AMG8833 from Panasonic, it’s cheap, easy enough to get, deliver a grid-eye style 8X8 thermal reading, and highly recognized for this type of project. In our opinion, it has the smallest form factor in its class.
We are using Raspberry Pi 4 (4GB Version) to process the main application, besides its compactness, no specific reason for choosing this Linux-based mini-pc, we happen to have this board laying around from the previous project.
The project uses:
- AMG8833 from Panasonic
- Raspberry Pi 4 (4GB Version)
- Lighting Tripod
- 5MP Camera Module
- 5Volt DC Fan
- X6 Project Box
- 3D Printed Holders
We split our engineers into two groups, one of them testing the limits of the thermal sensor and the options to read the data results, and the other group starts to code the face-recognition and to integrate it into one device. No one is underestimating the sensor capabilities, but at this point, we know that the 8X8 grid-eye is barely enough to do the job.
The prototype needs more than 100 hours to complete, and we manage to display thermal grid side by side with a standard view camera, we also add a person detection module. We name it thermal screen V0.
The device can handle 2 or 3 person detection simultaneously. The voice notifications work flawlessly; the device will tell us to wait for a moment for the screening process, giving a polite request to move forward if the object was not in range and even say thank you after the screening process is complete.
We achieved an almost 1-meter scanning distance, but the thermal detection result was not always consistent when the machine is running for hours. The 8X8 grid-eye thermal sensor seems to struggle to give its best performance, and to work with this sensor; the calibration process was quite tricky; it involves room temperature, ironing device (for calibration), thermal gun, and digital temp meter.
- The thermal result was not always consistent
- Performance drop after long-running
- Longer loading time (using Raspbian OS)
Second Chance does Exist
After a short discussion, we all agreed to proceed with the project to the next level. We have a better plan, this time, using a much higher performance sensor, a Malexis MLX90640, and an odroid tablet running a full android to enhance face & object recognition. This follow-up project uses:
- MLX90640 from Malexis
- 8" Odroid Tablet (Running full Android)
- Milled Aluminum Casing
The result was satisfactory for us. Higher performance thermal sensor making the device almost consumer-grade electronics. The next step is making the equipment working as an attendance system and integrate it with our Intelligent Operations Platform; to do that, our engineers create a user-friendly face recognition-based attendance system with multi-device synchronization capabilities. We call it Facely.
The attendance system can detect as many devices connected to it. The user can create a custom time interval to handles multiple working hours, allowing HR division to deliver a better solution in body temperature screening with a face recognition-based attendance system and promoting a hygienic lifestyle inside the company.
In conclusion, we are so surprised by the possibilities that such a little sensor could bring. If distance and reading speed is not a priority, the AMG8833 is doing the job very well, but since we need much faster performance, MLX90640 is the best thermal sensor we tested. We chose Android over Raspbian because face-recognition is natively available.