Signal, Image and Video Processing Обработка сигналов

Controllable iPPG video synthesis for enhanced pulse wave extraction and data augmentation

Zhang, Qi, Zhang, Zhenyin, Wu, Huayu, Li, Xinyang, Zhao, Xuemei
Springer Nature · 2026-11-16 · DOI 10.1007/s11760-026-05888-0

Аннотация

iPPG technology is one of the main methods for non-contact pulse wave measurement. However, a major challenge in current research focused on accurately extracting pulse waves using computer vision based on deep learning is the limited variety of pulse waves in existing datasets. Additionally, the videos or image sequences provided by these datasets are constrained by factors such as the camera’s frame rate and bit depth, ambient light during measurement, and the subject’s skin tone. These limitations hinder the model’s ability to learn how to extract pulse wave signals from videos. A method is proposed in this paper to synthesize videos containing pulse wave signals with different rates, shapes, and starting positions, while also allowing for the selection of desired skin tones and signal intensity in the RGB channels. This approach aims to expand datasets and enhance the model’s ability to extract pulse wave signals from videos. The method was evaluated using a pulse-presence discriminator, real–synthetic waveform comparisons, and a two-stage iPPG extractor training experiment. Parameter controllability and signal fidelity were further examined using 15 single-parameter conditions, 144 factorial configurations, and 96 additional configurations based on wrist skin backgrounds collected from two volunteers with darker skin tones. The results support controllable pulse-video synthesis and demonstrate improved waveform extraction and heart-rate estimation after synthetic-video continuation under the evaluated training configuration.

Читать у издателя →