This is one of the hottest applications to detect people's fever through image sensors. Artificial intelligence (AI) technology can improve the accuracy of body temperature detection. This article will show you the reference design of fever detection using the state-of-the-art CMOS thermal sensor and AI technology by Arrow Electronics, which will greatly improve the speed and accuracy of fever detection.
Using artificial intelligence to improve detection speed and accuracy
Since the outbreak of SARS in 2003, thermal imaging technology has gradually developed into a fever detection technology. At present, it has become an important tool to combat COVID-19 and is widely used in public places such as airports, stations and restaurants. However, with the rapid development of AI technology, the fever detection system will develop in a faster and more accurate direction.
Arrow Electronics has introduced a reference design using AI thermal sensing technology, aimed at fever detection applications. Using state-of-the-art CMOS thermal sensors, the sensor also uses ToF (Time of Flight) distance measurement sensor to compensate for temperature value. Through an RGB camera screen, thermal and RGB video can be displayed simultaneously. In this solution, Arrow Electronics has also jointly developed with product partners to carry out AI human body detection and human body core temperature calculation, and transplant them into an advanced ST Micro embedded processing unit.
Ultra-low noise improves detection accuracy
The hardware of the fever detection system solution consists of four main blocks. The first part is the main processor, which uses STMicroelectronics (ST) Arm-based Cortex MCU, a MCU with M7 and M4 dual-core processors (running at 480MHz). The second part uses a thermal imaging sensor with a resolution of 80 x 62 provided by Meridian Innovation, a new type of sensor fabricated in CMOS technology and specially designed to capture thermal energy. The third part is ST's 8 x 8 TOF sensor array distance detection system, a design using a system in package, which can integrate a laser diode and sensor into a single component of 6 x 3 mm size, and can simultaneously measure data in 64 zones. The fourth part is ST's digital ambient thermal sensor, which can convert actual temperature information into digital data and import it into the system.
In addition, for the power supply part, Arrow Electronics has also created a special design. In order to achieve better signal-to-noise ratio, ST's ultra-low noise low dropout regulator (LDO) is adopted, which can output noise lower than 6 micro RMS, so that clearer and more accurate images can be obtained. All three sensors are connected to the main processor through I2C and SPI interfaces, which will process all data processing and calculation.
Running artificial intelligence in embedded systems
For the software, Arrow Electronics and suppliers have developed a brand-new compensation algorithm, which uses the distance and environmental temperature information data collected by the system to calculate the core temperature of the human body; that is, the result of the system will not be affected by the distance and position of the sensor and detected object or the temperature condition of the installed location, thus generating a more stable and accurate detection data for the user.
In addition, there is a special thermal image noise reduction process executed in this system, which allows the system to receive clear and low noise thermal images. It is a special noise reduction process developed by the convolution neural network (CNN) system. This software was first developed and tested in a computer using TensorFlow, and then running on the MCU. In spite of 480 MHz speed and 2Mbytes memory, even less than one-tenth of the performance of a computer, the MCU can convert a TensorFlow Lite model into a special ST AI model through ST's latest STM32 Cube.AI intelligent software development tool, which can run on the MCU in real time, greatly reducing the system requirements for running artificial intelligence and machine learning framework in an embedded system.
Conclusion
At a time when the pandemic is affecting human life, fever detection equipment is the best tool to curb the spread. Through the new fever detection system, not only will the speed of fever detection be improved, but also the accuracy can be greatly improved. Arrow Electronics' AI fever detection reference design will shorten the development time of related products and enhance the market competitiveness of products, which is worthy of in-depth understanding by manufacturers interested in seizing the market.