Research Philosophy

I believe that a good way to do innovative research is to start with the basic concepts and first principles. Then working one's way to the top of a complicated problem, one can have a solid foundation. Thus, when a way to improve the state-of-the-art is unclear, elegant solutions can come from a different view of the basic concepts themselves. Being an engineer at heart, I also want to see a potential for immediate utility of my research. Thus, I keep an application in mind to give an overall direction to research.

Current Research

The target application that I have in mind is event detection in surveillance video. The problem itself comprises of many sub-tasks such as background subtraction, human detection, tracking, and event modeling. My current approach is to develop a paradigm for hierarchical modeling and processing of video. What this means is that sub-tasks at various levels will be handled by different modules inspired by the state-of-the-art methods in for respective sub-tasks, and interconnected in the hierarchical paradigm to work synergetically. So, the problem boils down to how the modules will interface with each other, what are the conditions favorable to synergetic learning between different learning modules, and what are the guaruntees that system perfomrance will improve with time. For real-time and complicated scenarios, online learning (as opposed to batch learning) will be preferred, which brings up the issue of optimal use of computational power, as more of it becomes available in the future. This hierarchical paradigm is inspired by graphical models such as Bayesian Networks and Factor Graphs, and the overall learning and inference is inspired by local message passing, and approximate, variational, and online versions of the EM-Algorithm, and Product of Experts.

This is the PhD thesis in progress.

Topics of Interest

Computer Vision and Machine Learning:
____Abnormal Event Detection in Video
____Hierarchical Models of Visual Processing Architecture and Information Representation
____Generative Models, Approximate Inference and Learning
____Computational Models of Perception and Cognition
____Probabilistic, Graphical, and Genertive Models for Video Understanding
____Background Subtraction
____Tracking rigid and non-rigid motion
____Scene Gist Classification in collaboration with Parallel Research in Psychology
____Structure from Motion from Smooth Textureless Objects using Silhouettes
____Object Recognition of Smooth Textureless Objects using Silhouettes

Genetic Algorithms:
____Walsh Schema Analysis of Simple Genetic Algorithms

Neural Network Applications:
____Robot Control
____Tree and Path lengths Optimization under constraints

Control, Robotics, and Mechatronics:
____Real-Time Manufacturing Process Control
____Self-Navigating Robots

Neuroscience and Perceptual Psychology:
____Scene Gist Recognition

Publications

(You will need Adobe Acrobat to view these publications in pdf format.)

"Variable Module Graphs: A Framework for Inference and Learning in Modular Vision Systems", A. Sethi, M. Rahurkar, T. S. Huang. International Conference on Image Processing (ICIP), 2005. [ pdf ]

"Using Visual Masking To Explore The Nature Of Scene Gist", L. Loschky, A. Sethi, D. J. Simons, D. Ochs, J. Corbielle, K. Gibb. Psychonomics (ICIP), 2005. [ pdf ]

"Robust Speaker Tracking by Fusion of Complementary Features from Audio and Video Modalities", M. Rahurkar, A. Sethi, T. S. Huang. Workshop on Image Analysis for Multimedia Interactive Services (WIAMIS), 2005. [ pdf ]

"Robust Structure and Motion from Outlines of Smooth Curved Surfaces", Y. Furukawa, A. Sethi, J. Ponce, D. J. Kriegman. Submitted to IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 2004. [ pdf ]

"A Detection-Based Multiple Object Tracking Method", M. Han, A. Sethi, Y. Gong. International Conference of Image Processing (ICIP), 2004. [ pdf ]

"Structure from Motion for Smooth Textureless Objects", Y. Furukawa, A. Sethi, J. Ponce, D. J. Kriegman. European Conference of Computer Vision (ECCV), 2004. [ pdf ]

"Curve and Surface Duals and the Recognition of Curved 3D Objects from their Silhouettes", A. Sethi, D. Renaudie, J. Ponce, D. J. Kriegman. International Journal of Computer Vision (IJCV), Vol. 58: No. 1, June 2004. [ pdf ]

"On Pencils of Tangent Planes and the Recognition of Smooth 3D shapes from Silhouettes", S. Lazebnik, A. Sethi, C. Schmid, D. J. Kriegman, J. Ponce, M. Hebert. European Conference of Computer Vision (ECCV), 2002. [ pdf ]

Associations at UIUC

Professor Thomas S. Huang, Image Formation and Processing Group, ECE, UIUC: 2002 - present

Professor David J. Kriegman, Artificial Intelligence Group, CS, UIUC (now at CS, UCSD): 2001 - 2002

Professor Henrique L.M. dos Reis, Nondestructive Testing and Evaluation Research Laboratory, GE, UIUC: 1999 - 2001

Professor Daniel Simons, Department of Psychology, UIUC: 2003 - 2004

Professor Lester Loschky, Department of Psychology, UIUC (now at Psych, KSU): 2003 - 2004

Resume