MSR Speaking Qualifier
December
Tuesday, December 6
12:00 pm to 1:00 pm
Gates Hillman Center 4405
MSR Thesis Talk: Ambareesh Revanur
Title: Towards Video-based Physiology Estimation
Abstract: RGB-video based human physiology estimation has a wide range of practical applications in telehealth, sports and deep fake detection. Therefore, researchers in the community have collected several video datasets and have advanced new methods over the years. In this dissertation, we study these methods extensively and aim to address the following limitations. First, several methods have exclusively focused on addressing the problem of heart rate estimation; but other vital-signs such as blood pressure have received little attention. Second, the datasets typically contain few subjects and record physiological data in resting conditions, which does not sufficiently represent patients with cardiovascular issues. Third, the evaluation protocol utilized by these methods does not account for fine-grained variations in physiological signals.
In this work, we introduce three new datasets each collected for a specific purpose. The Vision-4-Vitals (V4V) dataset was curated for heart rate estimation and respiration rate estimation using an emotion elicitation protocol to introduce variations in physiological state under laboratory settings. We are currently collecting another large scale dataset with over 2000 subjects at India and Sierra Leone sites for multiple physiological signal estimation including blood pressure and oxygen saturation. Furthermore, we are also in the process of collecting a new dual-camera based blood pressure dataset with baseline measurement and modified blood pressure measurements through breathing activities. Finally, we introduce the continuous evaluation protocol and a new Transformer based approach that outperforms existing heart rate and respiration rate estimation methods.
Committee:
Prof. László Jeni, Chair
Prof. Artur Dubrawski
Prof. Conrad Tucker
Mononito Goswami