Automatic Ingestion Monitoring (May 2018 - November 2019)

AIM is a an egocentric wearable sensor used to monitor dietary intake and ingestion behaviour. The project was funded by the National Institute of Diabetes and Digestive and Kidney Diseases. My research extended to addressing multiple issues related to the functionality of AIM and the Ingestion behaviour recognition.

  • Performing deblurring to restore motion blurry images and improving the battery life of the sensor by using a head orientation information to capture images only while a person is viewing the food.

  • Addressing privacy concerns related to content privacy and bystander privacy using deep convolutional neural networks to remove privacy content and accurately preserve food and environment-related content.

  • Developing a novel methods for ingestive behavior and environmental assessment to aid nutritionist.

  • Developing an automatic potion size estimation algorithm based on triangulation for the AIM-v2 camera images during food intake

  • Designing the computer vision aspects of the free-living ingestion monitoring study using the AIM

Ingestive behavior and environmental assessment

Develop recognition and assessment methods by applying deep learning recognition on the egocentric AIM images while leveraging information from IMU data of the AIM

Remote Heart Rate Estimation March 2015 - February 2018

H-Monitoring was a health monitoring prototype that measured vital signs such as: heart rate, respiratory rate and, autonomic nervous system functions using an non-contact RGB camera. H-Monitoring was a outcome of the deciertation "Non-Contact Heart Rate Monitoring Method by Maximizing The Quasi-Periodic Information of The RGB Camera Signal For Naturalistic Environment"

Python implementation of real-time remote heart rate estimation

Track3D March 2013 - December 2014

Track3D was a Visual surveillance based crowd analytic system. More particularly, relates to understanding and analyzing behavior of crowd at real time for dynamic environments. Track3D has various operations such as: illumination compensation, background modeling, ROI extraction, 3D tracking, motion estimation, density estimation and face recognition