BP-rPPG: An Indian Face-Video Dataset and PPG-Guided Baseline for Remote Blood Pressure Estimation

Abstract

Blood pressure (BP) is a vital health indicator, traditionally measured through contact-based methods, limiting remote applications. Remote photoplethysmography (rPPG) enables contactless BP estimation, yet its potential remains under-explored due to limited datasets. To address this, we introduce BP-rPPG, a challenging dataset for rPPG-based BP estimation that provides four video-quality variants to support systematic evaluation under varying compression conditions. Existing studies show that rPPG methods lag behind PPG-based ones in BP estimation, highlighting the need for improvement. Hence, we introduce ALIVE, a novel rPPG-based BP estimation method that extracts rPPG features resembling critical PPG features for BP estimation. To this end, ALIVE utilizes a specialized network architecture and custom loss function. Another issue in existing rPPG-based BP estimation methods is their dependence on entire facial videos and neglecting unevenly distributed rPPG information across frames. To address this, ALIVE introduces a novel mechanism, quality-weighed fusion, which segments videos into clips, extracts useful BP information for each clip, and eventually consolidates the extracted information for BP estimation. Experiments on BP-rPPG and the publicly available MSPM datasets show ALIVE outperforms existing methods. Also, the results highlight BP-rPPG as a more challenging dataset that aligns closely with real-world scenarios. The BP-rPPG dataset and the implementation of the ALIVE baseline are available at https://github.com/TrishnaSaikia/BP-rPPG-ALIVE.git. © 1975-2011 IEEE.