Camera-Augmented Non-Contact Vital Sign Monitoring in Real Time
The traditional method of monitoring the vital signs using sensors and wires attached to the body causes discomfort for patients. Researchers proposed a new system combining cameras and radars to address the limitations of traditional methods and successfully and effectively monitor vital signs in real-time.
The experimental setup utilized a Red Green Blue-Depth (RGB-D) camera and a Frequency Modulated Continuous Wave (FMCW) radar operating at 77-81 GHz. The radar employed a linear frequency modulated signal to detect subtle torso movements caused by the cardiopulmonary system. The radar estimated chest movement and provided valuable data for vital sign monitoring by processing the received signal.
The camera offered depth and RGB information, which was used to identify subjects' 3D locations via a computer vision algorithm. This algorithm employed convolutional neural networks to estimate body landmarks on the torso, aiding in subject movement tracking. Additionally, the radar adapted its beams based on camera input to monitor vital signs like heart rate and respiration rate.
Camera and radar coordinates were aligned to integrate data from both systems. Sparse uniform rectangular array (URA) configuration for the radar antennas was employed, enabling perception of movements associated with subjects at varying angles relative to the radar. The signal processing chain involved noise cancellation, phase extraction for chest movement waveforms, and vital sign estimation. Singular value-based point detection (SVPD) was used to locate the most representative points for vital sign extraction automatically.
Factors like distance, angle, and body landmarks that impacted the accuracy of vital sign measurements were investigated. It was inferred that closer distances and specific body landmarks led to more accurate measurements. Some signal interferences with two subjects, significantly when closer, were observed. However, more advanced radar structures were proposed to improve accuracy in such situations.
SVPD was tested on ten healthy individuals to assess its vital sign tracking accuracy. The experiment involved setting up a radar system transmitting chirp signals and a camera capturing visual data. These systems were synchronized for accurate vital sign measurement. Additionally, sensors were attached to subjects to obtain reference data for comparison. The experiments included scenarios where subjects sat still for two minutes, alone or with another person. Researchers measured variables like distance, angle, and region of interest to analyze their impact on the accuracy of vital signs.
The results indicated that SVPD achieved higher accuracy in measuring vital signs than traditional methods. For instance, it exhibited a 0.0% error in respiratory rate compared to a 68.21% error with the camera alone. Similarly, it showed a 2.41% error in heart rate compared to a 43.54% error with the camera alone. Compared to previous studies, SVPD outperformed many existing methods in terms of accuracy and processing time. The researchers also acknowledged limitations, including the need for improved camera systems with wider fields of view and more advanced radar structures for enhanced accuracy.
The study demonstrated the potential of combining radar and camera systems for accurate and real-time vital sign monitoring, holding significant promise for various applications, particularly in the healthcare domain.