Skip to main content

Video-based Remote Heart Rate Estimation

Primary supervisor

Munawar Hayat

The project aims to estimate heart rate non-intrusively using video data of humans faces. The heart rate estimation from videos finds its usefulness in numerous applications including patient contactless heart-rate monitoring in hospitals and elderly care facilities. The main challenges are associated with the subject's motion and varying lighting conditions. Existing methods mostly devise a multi-stage strategy, where the first stage detects facial region, performs tracking followed by segmentation of skin pixels. The extracted signal from the skin pixel is then used to extract changes or variations in "blood volume pulse". This project will aim to address some of the limitations of the existing techniques, and design/implement an end-to-end pipeline, which can estimate heart-rate under realistic conditions. For this purpose, deep Convolution Neural Networks (CNNs) will be explored, to predict the Blood Volume Pulse (BVP) signal.

Student cohort

Double Semester


- Developing a unified end-to-end deep learning pipeline to estimate instantaneous heart rate from persons facial video data

Required knowledge

- Deep Learning fundamentals

- Familiarization with Python, Pytorch, and Image/Video Processing Libraries

- Familizaration with basics of signal processing