Video Respiratory Rate Measurement in Walking Scenarios Using Multi-strategy Adaptive Denoising

For non-contact respiratory rate (RR) measurement, effectively addressing the interference from continuous motion artifacts remains a significant challenge. Most existing research focuses on the removal of weak motion artifacts in a two-dimensional plane, and the fixed spatial scale of the scenes limits the generalization of these methods to real-world scenarios, especially in real walking scenarios. To tackle this issue, we propose an RR measurement framework based on a multi-strategy fusion motion artifact suppression algorithm and have constructed a real-world walking dataset. Specifically, the framework consists of three core modules: an ROI automatic selection and adaptive enhancement module to guide the selection of high-quality corner points; a signal quality evaluation module that adaptively assesses whether the signal is noisy, preventing blind denoising; and a multi-strategy fusion motion artifact removal module that dynamically selects the appropriate strategy to suppress motion interference. To the best of our knowledge, this is the first study to investigate the task of video-based RR measurement in real walking scenarios. Experimental results demonstrate that the method achieves state-of-the-art performance across multiple datasets, with a mean absolute error (MAE) of 1.04 breaths per minute (bpm) on the COHFACE, 3.17 bpm on the OVRM-Walking dataset, and an average MAE of just 2.41 bpm on the in-house real-world walking dataset, which includes both indoor and outdoor scenarios. This study broadens the applicability of camera-based non-contact RR detection technology. The dataset and code will be publicly available in this project.