Research

This project will focus on addressing the following fundamental challenges key to a multitude of visual data analytics applications: 1) raw visual data cleaning; 2) visual data registration and fusion; and 3) visual data analytics and management. The existing efforts, either from the academia or the industry, are not capable of robustly and efficiently modeling, analyzing, and fusing continuous or discrete visual data captured by individuals or big companies.

Grounded on the recent novel and exciting developments described in this project, we plan to extend, generalize, and optimize them to address the aforementioned key challenges of visual modeling and analytics for the masses using the following research directions:

Localization – recovering the geometric locations of the user, the camera viewpoint, or the objects in the environment around the user;
Registration – aligning and modeling dynamically captured images and measurements of the scene over different time and viewpoints together;
Inference – estimating and analyzing the semantic information of the scene from the registered visual information and recovered geometric information.

We aim at achieving both high robustness and accuracy for the above tasks at unprecedented processing speeds on commodity computing devices and mobile cameras, often producing more than one or two orders of magnitude of speedup over the existing state-of-the-art solutions.

We have been working on the following clusters of research topics, and now are actively innovating in a broader scope.

New:

Our slides for a recent ADSC workshop on “Computer Vision for Autonomous Systems” [Part 1], [Part 2]
Recent talk slides presenting our work on “High-Quality Visual Correspondence Estimation with Simple, Generic and Efficient Techniques” (12.4M, 102 slides)