PPG-Based Heart Rate & Hydration Monitor

Analyzes photoplethysmography (PPG) signals from video to extract heart rate and assess hydration status using the CHROM method.

What I Built

Signal Extraction (CHROM Method):

• RGB channel processing to detect blood volume changes in skin
• Exploits differential absorption across color channels
• Processes video at 30 FPS

Heart Rate Detection:

• Bandpass filtering (0.7–4.0 Hz) for physiological range
• Derivative-based peak detection
• ±3 BPM accuracy vs. reference devices

Hydration Classification (TPA/VPA Ratio):

• TPA (Total Pulse Area): Area under pulse cycle
• VPA (Valley-to-Peak Area): Vascular compliance measure
• Thresholds: < 0.559 severe, 0.559-0.815 mild, 0.815-1.326 normal, > 1.326 overhydration

Firebase Integration:

• Video uploads to Firebase Storage
• Automatic processing on upload
• Results stored in Firestore with timestamps

Technical Details

CHROM Algorithm

Combines RGB channels to cancel motion artifacts:

1. Normalize R, G, B channels
2. Compute chrominance: Xs = 3R - 2G, Ys = 1.5R + G - 1.5B
3. Temporal filtering to remove DC
4. Isolate pulse: S = Xs - (σ(Xs)/σ(Ys)) × Ys

Pipeline

Video → Frame Extract → ROI (center) → RGB Extract → CHROM
  → Bandpass (0.7-4.0 Hz) → Peak Detect → HR + TPA/VPA → Classify

Optimizations

• Skip first/last 3 seconds (initialization artifacts)
• Vectorized NumPy operations
• Memory-efficient streaming (no full video load)
• GPU acceleration option for real-time

CLI

# Single video
python -m src.main video.mov --save-trace trace.png

# Batch processing
python -m src.main *.mov --output results.json

# Live visualization
python -m src.main video.mov --show-plot

Results

Performance:

• Heart rate: ±3 BPM accuracy
• Processing: ~10s for 30s video (M1 MacBook Pro)
• TPA/VPA correlates with urine specific gravity (validation pending)

Key Insights:

• Lighting critical: consistent, diffuse lighting improves signal
• Skin tone requires calibrated CHROM coefficients
• Motion > 5mm degrades signal—face tracking would help
• TPA/VPA ratio novel (not in commercial devices)—potential publication

Contactless Advantage:

• Remote patient monitoring without contact
• Mass screening via webcam
• Low-cost alternative to wearables

Future Work

• Face tracking for motion compensation
• ML model to predict hydration from waveform features
• SpO2 estimation via R/IR ratio
• Web interface for upload/analysis
• Clinical validation study