Heart Rate Estimation using Hermite Transform Video Magnification and Deep Learning^.

Area of Science:

• Biomedical Engineering
• Computer Vision
• Signal Processing

Background:

• Traditional heart rate monitoring faces portability and connectivity challenges outside clinical settings.
• Computer vision offers non-contact measurement of physiological variables like heart rate.
• Video magnification techniques, such as using a Gaussian pyramid, aid in pulse signal detection.

Purpose of the Study:

• To propose a novel strategy for motion magnification in video sequences using the Hermite transform.
• To implement a deep learning technique for estimating beat-by-beat pulse signals.
• To compare the proposed method with classical Gaussian pyramid video magnification for heart rate monitoring.

Main Methods:

• Utilized the Hermite transform for motion magnification in video sequences.
• Implemented a deep learning model for beat-by-beat pulse signal estimation.
• Validated results against an electronic pulse monitoring device and compared with Gaussian pyramid methods.

Main Results:

• The Hermite transform-based approach demonstrated superior enhancement of spectral information from color changes.
• The deep learning technique, combined with Hermite transform magnification, enabled accurate instantaneous beat-by-beat pulse estimation.
• The proposed method outperformed the classical Gaussian pyramid approach in enhancing relevant spectral information.

Conclusions:

• The novel Hermite transform and deep learning strategy provides an accurate and enhanced method for non-contact, beat-by-beat heart rate monitoring.
• This approach overcomes limitations of traditional monitoring devices, offering greater portability and ease of use.
• The study highlights the potential of advanced video magnification and deep learning for remote physiological monitoring applications.