Research

Publications

I enjoy doing research and being able to share results of our work. I’ve been blessed to work with some amazing researchers over the years and to able to present the work at some great conferences. However for a full and up-to-date list of all my publications please visit my Google Scholar page.

Thesis

1. 3D Face Reconstruction from Monocular Video and its Applications In the Wild

   2020

   Committee: Laszlo A. Jeni (Co-Chair), Jeffrey F. Cohn (Co-Chair), Louis-Philippe Morency, Chaoyang Wang

   3D face reconstruction is a very popular field of computer vision due to its applications in social media, entertainment and health. However, ever since the introduction of 3D morphable models as facial priors, 3D face reconstruction has been dominated by reconstruction from single images due to its ease and proximity to 3D face alignment. Even so, single image reconstruction methods suffer from inconsistent reconstructions across time and view points. Hence a natural extension is to reconstruct 3D face shape from videos. Because of recent methods in single image reconstruction setting the standards for state-of-the-art reconstruction, we introduce a method to fuse single image reconstructions across multiple frames to create a more accurate reconstruction. Furthermore, the lack of structured video datasets that fully captures the face and provide 3D ground truth scans, led us to develop and release the 3DFAW-Video dataset and challenge. We also introduce a symmetric distance metric for benchmarking reconstructions on the 3DFAW-Video dataset that is less affected by reconstruction density. Finally, we discuss the usage of 3D face reconstructions in two different applications to be deployed’in-the-wild’. In particular, we illustrate applications in mask-sizing from a metric face 3D reconstruction, and present 3D face normalization as a technique to improve vision based non-contact heart rate estimation methods.

   See Thesis
2. Mixed initiative survivable path planning using imprecise information

   2018

   Advisor: Gianni Di Caro

   Safe robot navigation in human-robot teams during post-disaster scenarios such as after an earthquake, is a challenge for robots due to the cluttered, potentially un- known and hazardous environment. However, the human teammate’s advanced cognitive abilities can be exploited to aid in safe robot navigation in the difficult to navigate environment. This research aims to create a mixed initiative path planning system that receives the human’s advisory information that models potentially un- detected hazards in the environment and integrates it to the robots world view to plan paths that are the most survivable - a path with low danger and high probability of existence. Current research mostly focuses on solely autonomous planning using bayesian methods for map creation and updation, and uses precise spatial in- formation through gestures and speech to provide human input to robots. However, we consider human input that is inherently high-level and imprecise due to the use of ambiguous language. Additionally we also use imprecise probabilities for map creation and updation instead of bayesian probabilities due to difficulties in having a generalizable model of the "human sensor" that can refine our estimates over time. We discuss a method that uses a novel map representation based on imprecise prob- abilities, integrates imprecise human input modeling potentially undetected hazard to robot’s world map, and plans survivable and efficient paths using a modified rapidly exploring random tree (RRT) algorithm.

   See Thesis