Oracle headroom without signal: null-calibrated evaluation of candidate selection for thermal heart rate estimation

Mohammad Rakibur Rahman; Nhi Nguyen; Sasan Sharifipour; Le Nguyen; Manuel Lage Cañellas; Miguel Bordallo López; Constantino Álvarez Casado

cs.CVOct 1, 2026

Oracle headroom without signal: null-calibrated evaluation of candidate selection for thermal heart rate estimation

Authors: Mohammad Rakibur Rahman, Nhi Nguyen, Sasan Sharifipour, Le Nguyen, Manuel Lage Cañellas, Miguel Bordallo López, Constantino Álvarez Casado

Organizations: Center for Machine Vision and Signal Analysis (CMVS), University of Oulu, Finland

Abstract

Camera-based physiological monitoring can produce multiple estimates from several facial regions, extraction methods, and processing settings. Signal quality indices aim to select reliable estimates without a physiological reference, and their potential is often assessed with an oracle that selects the estimate closest to the reference in each window. This retrospective selection can reward chance agreement. We model the effect with order statistics. For K independent candidates unrelated to the reference, the expected oracle error decreases approximately as 1/K. We analyze thermal heart rate estimation on 96 iBVP recordings with 168 candidates per 10 s window. The oracle achieves a mean absolute error of 0.91 bpm, compared with 10.74 bpm for the best fixed configuration, 18.03 bpm for the best quality index, and 8.61 bpm for a constant predictor. With K = 24, a forehead signal from another recording matches the correct one, with 4.62 against 4.61 bpm. Oracle evaluations should report candidate count, valid coverage, and matched null controls.

Figures & tables

Figure 1: Null-calibrated oracle evaluation. A thermal crop from iBVP (recording p02a, $72\times 72$ pixels) with the six fixed regions of interest (ROIs) used here, and the learned model output, give seven signals. Each passes through 24 processing settings. Fixed and quality selection use signals only. The oracle also uses the reference. Matched null inputs pass through the same settings, so that oracle errors can be compared at equal candidate count $K$ .

Selection rule	Reference use	MAE (bpm)
Constant, 74.6 bpm	Mean over data	8.61
Best fixed setting	One choice	10.74
Best SQI (zero-crossing rate)	Index choice	18.03
Oracle, $K=168$	Every window	0.91
Uniform guesses, $K=168$	None	0.40

Table 1: Selection rules on the thermal bank of 168 candidates (6,816 windows). The last column states how the reference is used. The uniform row is the model of Eq. ( 3 ), not a measurement. Lower MAE is better.

Panel of Figure 2 — Figure 2: Window rejection applied to the best fixed setting (10.74 bpm at full coverage). Windows are discarded at random, by the SQI ranking of the best index, or by true error. Only rejection that uses the reference reduces the MAE.

Panel of Figure 3 — Figure 3: Oracle and fixed MAE against candidate count. Filled markers are means over 300 random subsets of the bank of 168 candidates (single evaluation at $K=168$ ). The dash-dotted curve is Eq. ( 3 ) for independent guesses uniform on 45 to 180 bpm. Markers at $K=24$ are banks from one signal in Table 2 .

Signal	Keeps	Fixed	Oracle
White noise	Nothing	17.91	11.34
Shuffled forehead	Amplitudes	18.02	11.05
Pink noise	$1/f$ spectrum	12.00	6.92
Donor forehead	Thermal dynamics	10.62	4.62
Forehead	All	10.74	4.61
Perinasal	All	10.99	4.19

Table 2: Matched null ladder with

K=24

settings per signal. Fixed and oracle MAEs are in bpm. The second column lists what each signal keeps from real thermal data. Lower is better.

Check (data)	Result
Training targets, 12 clips	PPG peaks at 62.4 to 92.3 bpm, 81 to 99% band power
16-bit frames, 4 rec.	Thermal peak 45.0 to 52.1 bpm, reference 69.9 to 85.5 bpm
Six ROIs, 96 rec., 20 s	$r$ from $-0.200$ to $0.023$ , MAE 21.1 to 24.2 bpm, constant 8.51 bpm
Lags of $\pm 30$ s, 96 rec.	Best $r$ 0.003 to 0.052, donor 0.013 to 0.049
OMIT [ 1 ] , 96 rec.	15.24 bpm at 30 Hz, 14.75 bpm at 7.5 Hz, $r\approx-0.05$
EfficientPhys, 8 folds	Validation loss near 1.0 from the first epoch

Table 3: Diagnostics of the thermal signal chain. Protocols differ from Tables 1 and 2 . Rate errors are MAEs, and

r

is the Pearson correlation.

Explore similar work

Thermal Imaging for Contactless Cardiorespiratory and Sudomotor Response Monitoring
Feb 12, 2026Constantino Álvarez Casado, Mohammad Rahman, Sasan Sharifipour +4Thermal Imaging Heart Rate Estimation
Center for Machine Vision and Signal Analysis (CMVS), University of Oulu, Finland · Candour Ltd, Oulu, Finland
Learning to Assess Heartbeat Observability for mmWave Heart-Rate Sensing
Oct 2, 2026Yuxuan Hu, Shilin Shan, Jianfei Yang +1Heart Rate Estimation mmWave Radar Sensing
Key Laboratory for Information Science of Electromagnetic Waves, Ministry of Education, School of Information Science and Technology, Fudan University, Shanghai 200433, China · School of Mechanical and Aerospace Engineering, Nanyang Technological University, Singapore 639798
Deep Learning for Sleep Heart Rate Estimation from Accelerometers: Toward Population-Scale Cardiac Insight Without Optical Sensors
Oct 5, 2026Tanbin Islam Rohan, Pranjol Sen Gupta, Tanusree Debi +1Transformer Wearable Sensing
Department of Information Technology Kennesaw State University Kennesaw, GA, USA · School of Computing University of Georgia Athens, GA, USA

Feb 12, 2026J/K move · Enter open · S save

Thermal Imaging for Contactless Cardiorespiratory and Sudomotor Response Monitoring

Constantino Álvarez Casado, Mohammad Rahman, Sasan Sharifipour, Nhi Nguyen+3

Center for Machine Vision and Signal Analysis (CMVS), University of Oulu, Finland · Candour Ltd, Oulu, Finland

Thermal Imaging Heart Rate Estimation Driver Monitoring

Open paper AskPDF arXiv

Human-machine interfaces in industrial automation need sensing modules that monitor operator actions and physiological state. This is important in factories, vehicles, machinery cabins, and human-robot collaboration, where workload, stress, fatigue, or reduced attention can affect safety. RGB monitoring is limited by low light, shadows, and privacy concerns, while thermal infrared imaging captures skin temperature dynamics without visible illumination. This paper studies thermal video as a contactless computer vision modality for estimating electrodermal activity (EDA), heart rate (HR), and breathing rate (BR), with the goal of supporting adaptive human-machine interfaces and operator-state awareness. We propose a signal-processing pipeline that tracks facial regions, aggregates thermal signals, and separates slow sudomotor trends from faster cardiorespiratory components. HR is estimated using orthogonal matrix image transformation (OMIT) across multiple facial regions, while BR is estimated from nasal and cheek thermal signals using spectral peak detection. We characterize 288 ROI-method configurations against contact references with lag-tolerant metrics using 31 sessions from the public SIMULATOR STUDY 1 (SIM1) driver monitoring dataset. The best fixed EDA configuration reaches a mean absolute correlation of

0.40 \pm 0.23

against palm EDA, with individual sessions reaching

0.89

. BR estimation achieves

3.1 \pm 1.1

bpm mean absolute error, while HR estimation yields

13.8 \pm 7.5

bpm MAE, limited by the

7.5

Hz thermal camera frame rate. The results show that thermal video provides useful respiratory and sudomotor cues, while revealing limitations caused by ROI selection, polarity changes, latency, and subject variability. These findings provide baseline design guidance for thermal computer vision as an auxiliary sensing layer in adaptive industrial HMI systems.

Abstract

Explore similar work

Thermal Imaging for Contactless Cardiorespiratory and Sudomotor Response Monitoring

Learning to Assess Heartbeat Observability for mmWave Heart-Rate Sensing

Deep Learning for Sleep Heart Rate Estimation from Accelerometers: Toward Population-Scale Cardiac Insight Without Optical Sensors