Conceptos Básicos
A novel brain-inspired framework for robust remote heart rate measurement that overcomes challenges of environmental influences, subject movement, and privacy concerns by extracting skin ROI without relying on facial detection.
Resumen
The paper proposes a novel remote heart rate measurement framework called HR-RST that consists of three phases: ROI extraction, signal analysis, and heart rate calculation.
ROI Extraction Phase:
- Utilizes a brain-inspired neural network model called CCNN to encode changing pixels (skin) into chaotic signals and static pixels (background) into periodic signals, enabling robust skin ROI extraction.
- CCNN-based ROI extraction overcomes challenges of face detection failure due to head movements and enables extending heart rate measurement to other body parts beyond the face.
Signal Analysis Phase:
- Performs spatiotemporal feature analysis of optical flow signals within the ROI video.
- Extracts RGB signals, filters the G-channel, and conducts time-frequency analysis on temporal signals for heart rate calculation.
Heart Rate Calculation Phase:
- Calculates the heart rate by finding the mode value of the heart rate distribution across all ROI pixels.
The proposed framework addresses the limitations of existing methods, including inaccuracies in face detection, applicability issues for special patients, and privacy concerns. It demonstrates robustness against motion and enables reliable ROI extraction, extending the applicability to other body parts beyond the face.
Estadísticas
The heart rate ground truth (HRgt) is 65 bpm.
The estimated heart rate (HRes) for different body parts are:
Palm: 43 bpm
Back of hand: 30 bpm
Forearm: 43 bpm
Upper arm: 43 bpm
Back: 43 bpm
Sole: 43 bpm
The heart rate error (HRer = HRes - HRgt) ranges from -22 bpm to -35 bpm across the different body parts.