Skip to main content

​
Overview

When a Muse headband with PPG is available alongside camera-based rPPG, the SDK supports sensor fusion — using the contact PPG as a reference to calibrate and improve camera-based heart rate estimates.

​
MuseFusionCalibrator

Fuses Muse PPG readings with camera rPPG estimates for improved accuracy: 
import { RppgProcessor, MuseFusionCalibrator } from "@elata-biosciences/rppg-web";

const processor = new RppgProcessor("wasm", 30);
const calibrator = new MuseFusionCalibrator();

// Feed Muse PPG readings (from headband)
calibrator.updateMuse(museBpm, quality, timestampMs);

// Feed camera rPPG readings
calibrator.updateCamera(cameraBpm, cameraQuality, timestampMs);

// Get fused result
const fused = calibrator.fuse(cameraBpm, cameraQuality, timestampMs);
The RppgProcessor also has a built-in Muse integration: 
processor.updateMuseMetrics(museBpm, quality, timestampMs);

​
MuseCalibrationModel

A simple regression model that learns the relationship between camera and contact BPM over time: 
import { MuseCalibrationModel } from "@elata-biosciences/rppg-web";

const model = new MuseCalibrationModel();

// Train with paired observations
model.train(spectralBpm, acfBpm, trueBpm);

// Check if enough data collected
if (model.isTrained()) {
  const predictedBpm = model.predict(spectralBpm, acfBpm);
}

// Persistence
const snapshot = model.getSnapshot();
model.loadSnapshot(snapshot);
model.reset();

​
BPM Evidence Types

The processor produces evidence from multiple analysis methods: 
type BpmEvidenceSource = "spectral" | "acf" | "tracker" | "muse" | "calibrated";

type BpmEvidence = {
  source: BpmEvidenceSource;
  bpm: number;
  quality: number;
  timestampMs: number;
};

type BpmResolutionResult = {
  bpm: number;
  quality: number;
  source: BpmEvidenceSource;
  evidence: BpmEvidence[];
};

type FusionSource = "camera" | "muse" | "fused";

​
Muse PPG Filter

Apply Muse-style bandpass filtering to PPG samples: 
import { museStyleFilter } from "@elata-biosciences/rppg-web";

const filtered = museStyleFilter(ppgSamples, sampleRate);

​
Calibration Workflow

  1. Start with camera-only rPPG
  2. When a Muse headband connects, feed its PPG as ground truth
  3. The calibration model learns the camera-to-contact mapping
  4. Once trained, camera-only estimates are corrected using the learned model
  5. The fusion calibrator combines both sources for maximum accuracy