Vineet R. Shenoy

I am a doctoral student at Johns Hopkins University and work with Professor Rama Chellappa. My research aims are to build algorithms for non-contact vital sign estimation using machine learning and optimization. I received my bachelors degrees in Electrical and Computer Engineering as well as Computer Science from Rutgers Unversity-New Brunswick

Email  /  CV  /  Google Scholar  /  Github

I am broadly interested remote physiological monitoring using cameras, with a particular focus on imaging Photoplethysmography (iPPG). I have developed algorithms for lab-based iPPG, and more recently, built algorithms for clinical applications and real-world settings.

We use generative models to learn the distribution of pulse signals given video, allowing for uncertainty quantification.

We apply iPPG techniques to hands with acute trauma for perfusion assessment in-field.

We obtain iPPG estimates from facial video by solving an inverse problem using deep equilibrium models.

We build a UNet model for imaging Photoplethysmography that achieves state of the art results for data in both the near-infrared domain as well as the RGB domain.

We generate pseudo-samples from low-data classes and learn a robust model using contrastive learning for the task of Few-Shot Object Detection.

We obtain the Blood Volume Pulse (BVP) wave from facial video by solving an inverse problem with learnable priors.

We develop a real-time multi-camera tracking system that works on operational camera data. We integrate the system into the RITIS platform and evaluate our algorithms on the AICITY 2021 Multi-Camera Tracking dataset.

We classify vehicle actions as such as "right turn", "left turn" and "straight through" by solving a sixth-order polynomial that depends on the vehicle's trajectory and the parametric definition of a movement.

As a part of the NVIDIA AI Challenge, we develop robust algorithms for multi-camera tracking, vehicle re-identification, and anomaly detection. We are among the top scoring teams on the public leaderboard.

As higher priority spectrum users request access to the 3.7 GHz band, incumbent users must determine when to vacate the band and when to request more access. We study timing constraints so that the channel is used efficiently and all FCC timing requirements are met.