With the continuous improvement of the quality of life, people are paying more and more attention to their own health status, so the demand for healthcare products is getting higher and higher, which provides great opportunities for the healthcare market. In particular, respiratory monitoring applications are one of the important functional requirements of elderly care and infant care, because these people especially need additional care to monitor their health status.
A non-contact breathing detection device is suitable for long period monitoring
At present, there are various breath detection devices on the market, e.g. ECG monitors or wearable devices. However, these devices usually require connection with wires and electrodes, or the devices themselves must come into contact with the human body for measurement. These detection devices may cause burden, discomfort, and even danger to users, so they are not suitable for long period monitoring.
Non-contact or non-invasive breath detection devices are suitable for the long period detection of respiratory rate and measurement of newborn babies. Using an RGB camera is one of the feasible solutions for breath detection, because RGB cameras are low-cost and widely available, and unlike wireless wearable devices, they have no battery life problem and will not be adversely affected by unknown RF.
Breath detection with a camera faces challenges
Although it is convenient to use an RGB camera for breath detection, there are still several challenges with such breath detection devices. For example, the human body's breathing movements do not fluctuate much, which may cause the RGB camera to detect only a few pixel offsets when breathing movements are detected. The computer may not be able to easily detect such subtle movements. Moreover, any slight movement of the human body may overwhelm the slight movement amplitude of breathing, so the measured human body must keep calm and stable, for example, in a sleeping state, in order to obtain the best signal-to-noise ratio. In addition, any movement in the background will also generate noise, thus affecting the accuracy of breathing measurement.
Moreover, the environment in which the frame is captured also affects the accuracy of breath detection. For example, if you are in an environment with insufficient light, since the camera must capture the frame with higher ISO sensitivity, higher signal noise will be generated. In addition, if the subject is wearing plain or dark clothes, the camera may not be able to see obvious breathing movement, and if the subject is wearing clothes with patterns or coarse textures, better measurement results can be provided.
Breathing detection algorithm improves detection accuracy
In order to improve the above problems, Arrow Electronics, in cooperation with a local institute in Hong Kong, has developed an artificial intelligence (AI) algorithm for breath detection, which is specially designed for edge devices such as NXP i.MX8M mini. The breath detection API was developed on the IMX8MMEVK evaluation kit. The key component of this algorithm is the Eulerian Motion Magnification filter. This temporal image filter magnifies the subtle motion in the video stream captured by the RGB camera. By adjusting the filter, subtle breathing movements can be enlarged to obtain the best respiratory rate measurement results. Then, the motion magnified image is compared with the original frame to extract motion features and translated them into a score. Then, the score is compared with the dynamic threshold to perform respiratory counting.
To prevent false detection, the algorithm also includes a mechanism that can be used to detect the status of the video stream to ensure that the frame captured by the camera can already be used for breath detection. With the help of AI technology, TensorFlow lite is used to detect the human body in the frame of the camera, and the detected human body will be outlined for breath detection so that any other motion in the background can be isolated.
The breath detection algorithm provides a simple and effective mechanism for developers to adapt API functions to user applications. After the API is initialized, the user application will push the camera frame to the breath detection API, which will perform all image processing and calculation. The event-based reporting system will regularly report the detected respiratory rate and feedback the updated API status for further analysis.
In order to evaluate the measurement accuracy of the breath detection algorithm, in the test, the camera is facing different body portrait at the front and the side for detection, and the data measured by the ECG monitor is compared at the same time. After detection in different seconds, the test results show that the algorithm has only a ±1 breath difference per minute when measuring different body portrait.
Evaluation board using chips of the moment’s choice
In terms of hardware, for basic applications, the IMX8M Mini or Nano can be used as the core of the system. For a complete system, additional flash memory, DDR RAM, power supply, and image sensors are needed. I.MX8M also provides different IP blocks to extend system functions, such as wireless communication and multimedia input and output functions.
The chips recommended for the evaluation board are the moment’s choice. Specifically, MIMX8MN or MIMX8MM from NXP is recommended for the main processor. The image sensor used is MT9M114 from ON Semiconductor. The power management IC is PCA9450 from NXP. The digital microphone used is MP34DT05-A from STMicro or SD07OT261 from Goertek. The audio codec used is AW8896 from Awinic or ADAU1361 from ADI. The DDR4 RAM used is MT40A1G16 from Micron or IS43QR16256B from ISSI. The EMMC flash memory used is MTFC2G from Micron. The dual-mode Wi-Fi & Bluetooth chipset used is CWY43439 from Infineon or 88W8987 from NXP, or Murata Type 1ZM dual-mode Wi-Fi & Bluetooth module. The crystal used is XRCGB24M from Murata. The HDMI transmitter used is ADV7535 from ADI. The USB PD PHY used is PTN5110NHQZ from NXP.
Conclusion
In cooperation with a local institute, Arrow Electronics provides this breath detection solution which uses a low-cost RGB camera on the I.MX8M Mini platform and has developed a non-contact breath detection API. Its key components include the Eulerian Motion Magnification filter, which can magnify subtle breathing movements. With the help of AI technology, breath detection API provides excellent performance suitable for healthcare applications such as human breath detection. The simplicity of the breath detection API design will help application developers to easily deploy breath detection features to their applications, quickly develop breath detection device products, and seize market opportunities.