Imaging Heart Rate Detection Method Based on Clustering and Adaptive Filtering

Fig. 2. Schematics of concave lens deformation algorithm before and after processing. (a) Before processing; (b) after processing; (c) example of the effect diagram after actual processing

Download full size

Fig. 3. Curves of NLMS algorithm simulation data. (a) Ideal sinusoidal signal $y (n)$ ; (b) output signal $d (n)$ after adding noise signal $X_{I R}$ and $y (n)$ ; (c) output curve by traditional NLMS denoised algorithm; (d) output curve by self-adaptive NLMS denoised algorithm

Download full size

Fig. 4. Background light intensity curve in the real-world scene

Download full size

Fig. 5. An example of the experimental scenario

Download full size

Fig. 6. Bland-Altman plots of different methods for subjects in UBFC-rPPG database. (a) Bland-Altman plot obtained by CHROM algorithm; (b) Bland-Altman plot obtained by EEMD algorithm; (c) Bland-Altman plot obtained by POS algorithm; (d) Bland-Altman plot obtained by NLMS algorithm; (e) Bland-Altman plot obtained by our method

Download full size

Fig. 7. Bland-Altman plots of the subject in a scene with dramatic changes in illumination. (a) Bland-Altman plot obtained by EEMD algorithm; (b) Bland-Altman plot obtained by POS algorithm; (c) Bland-Altman plot obtained by CHROM algorithm; (d) Bland-Altman plot obtained by traditional NLMS algorithm; (e) Bland-Altman plot obtained by improved NLMS algorithm; (f) Bland-Altman plot obtained by combining CHROM with the traditional NLMS algorithm; (g) Bland-Altman plot obtained by our method

Download full size

Table 1. Ablation experiment results on the UBFC-rPPG dataset

View table

Table 1. Ablation experiment results on the UBFC-rPPG dataset

Method	MAE /min^-1	MAPE /%	SD /min^-1	r
Original video	35.38	35.60	3.78	-0.01
Original video + concave lens deformation	11.97	11.99	2.95	0.58
Original video + SKIN	36.70	36.82	4.05	-0.06
Original video + SKIN + K-means	5.11	5.60	2.97	0.66
Original video + concave lens deformation + SKIN + K-means	4.29	4.19	2.59	0.66

Table 2. Results of performance comparison experiments in the head motion scenario on UBFC-rPPG dataset

View table

Table 2. Results of performance comparison experiments in the head motion scenario on UBFC-rPPG dataset

Approach	Method	MAE /min^-1	MAPE /%	SD /min^-1	r
Unsupervised signal processing methods	Ours	0.92	1.57	2.43	0.65
	CHROM^［5］（2013）	4.29	4.19	2.59	0.66
	EEMD^［15］（2015）	30.06	31.01	2.88	0.16
	POS^［12］（2017）	1.49	1.60	2.44	0.74
	NLMS^［14］（2018）	8.42	9.97	2.98	0.42
Supervised deep learning methods	Meta-rPPG^［16］（2020）	5.97	—	7.12	0.53
	AND-rPPG^［17］（2022）	2.67	—	3.21	0.96
	X-iPPGNet^［18］（2023）	4.99	—	6.25	0.67

Table 3. Comparison of measurement results for different methods in scenes with severe lighting changes

View table

Table 3. Comparison of measurement results for different methods in scenes with severe lighting changes

Method	MAE /min^-1	MAPE /%	SD /min^-1	r
EEMD^［15］	21.93	26.70	2.28	-0.12
POS^［12］	11.14	12.90	2.88	0.23
CHROM^［5］	15.35	17.48	3.07	0.05
NLMS^［14］	18.15	23.25	3.52	0.01
NLMS（Ours）	14.17	18.30	3.28	0.03
CHROM+NLMS	8.79	11.21	3.14	0.13
Ours	2.20	3.68	2.55	0.23

Tools

Get Citation

Copy Citation Text

Manping Huang, Li Peng, Peng Han, Kaiqing Luo, Dongmei Liu, Miao Chen, Jian Qiu. Imaging Heart Rate Detection Method Based on Clustering and Adaptive Filtering[J]. Acta Optica Sinica, 2024, 44(9): 0917002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Medical optics and biotechnology

Received: Jan. 2, 2024

Accepted: Feb. 22, 2024

Published Online: May. 15, 2024

The Author Email: Li Peng (qiuj@scnu.edu.cn), Jian Qiu (pengli@m.scnu.edu.cn)

DOI:10.3788/AOS240433

CSTR:32393.14.AOS240433

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology