Fast And Robust Heart Rate Estimation From Videos Through Dynamic Region Selection

Area of Science:

• Biomedical Engineering
• Computer Vision
• Signal Processing

Background:

• Remote heart rate (HR) estimation from videos offers a non-contact method for health monitoring.
• Accurate HR estimation relies heavily on selecting appropriate image regions containing clear pulse signals.
• Existing methods often struggle with accuracy due to static region selection or environmental variations.

Purpose of the Study:

• To propose a novel algorithm for accurate remote heart rate estimation from videos.
• To enhance the robustness and efficiency of HR estimation through dynamic region selection and advanced tracking.
• To provide a sensor-free solution for continuous health condition monitoring.

Main Methods:

• A machine learning-based region selector identifies image areas with discernible pulse waveforms.
• A particle filter is employed to robustly track temporal HR variations.
• The algorithm dynamically selects regions for HR calculation, optimizing accuracy.

Main Results:

• The proposed method achieves an absolute average error of less than 1.1 Beats Per Minute (BPM).
• Processing time for a single HR estimation is under 0.6 seconds.
• Demonstrated superior accuracy and robustness compared to traditional methods.

Conclusions:

• Dynamic region selection significantly improves the accuracy of remote HR estimation from videos.
• The integration of machine learning and particle filtering offers a robust and efficient solution for non-contact HR monitoring.
• This sensor-free approach has potential applications in remote healthcare and wellness tracking.