A research tool that estimates core body temperature in real-time by applying a custom Extended Kalman Filter (EKF) to heart rate data streamed from a Polar heart rate monitor — a non-invasive alternative to clinical thermometry, built in collaboration with Dr. Udayraj.
Role
Algorithm Engineer & Research Developer
Year
2024
Category
Health Tech / Research
Tech Stack
The problem
Note · 01Measuring core body temperature accurately requires invasive probes or expensive infrared equipment. The challenge was to build a reliable, non-invasive estimation pipeline using only heart rate — a signal that has a known physiological relationship with thermoregulation but is highly noisy and context-dependent.
The approach
Note · 02Designed a custom Extended Kalman Filter that models the nonlinear relationship between heart rate and core temperature, with separate exercise and recovery phase models. Heart rate data is streamed live from a Polar monitor via the Google Fit API, minute-averaged to reduce noise, and fed into the EKF pipeline. The filter dynamically switches between quadratic models based on delta-HR direction, producing smooth, calibrated core temperature estimates.
How it was built
Milestones from concept through delivery.
Reviewed literature on the physiological coupling between heart rate variability and thermoregulation
Prototyped the EKF model in MATLAB (Proposed_EKF_model.m) to validate the filter equations and coefficient fitting on collected exercise data.
Implemented the full pipeline in Python
Built the EKF core (ekf_model.py / heartrate_2.py) with dual
exercise model (delta-HR ≥ 0) and recovery model (delta-HR < 0), each with tuned quadratic coefficients.
Validated estimated core temperatures against clinical thermometer readings across varied activity intensities with collaborator Sai Kamal Nannuri.
Impact
Visuals
Live ECG trace captured from the Polar heart rate monitor — the raw input to the EKF core temperature estimation pipeline.
Real-time core temperature estimates rendered on screen — the Extended Kalman Filter output converging to clinically relevant values.
