Atul Kanaujia
23960 BPO Way,
Piscataway, NJ 08854.
Ph.: +1-732-762-4888
http://www.cs.rutgers.edu/~kanaujia
RESEARCH INTERESTS
Human Pose Reconstruction, Human Activity Recognition, Manifold Learning, Facial Features Tracking, Facial Expressions Analysis.
EDUCATION
PhD. in Computer Science September 2007 (Expected)
Rutgers University, Piscataway, NJ
Thesis: Conditional Models for 3D Human Pose Reconstruction
Advisor: Dimitris N. Metaxas
Masters of Science in Computer Science October 2003
Rutgers University, Piscataway, NJ
GPA: 3.72/4.00
Bachelors of Science in Computer Science and Engineering May 2000
Indian Institute of Technology(IIT), Bombay
GPA: 8.28/10.00
Advisor: Dr. S. Biswas
RESEARCH EXPERIENCE
Research Assist. Comp. Biomedicine Imaging and Modeling Lab Jan 2004 - Current
Learning to Reconstruct 3D human motion from 2D monocular Images
Research focused on motion analysis and 3D pose reconstruction of complex articulated structure of human body from monocular 2D images. Perspective projection of 3D human body gives rise to ambiguities with uncertain perception inferences that cannot be learned using single regressor. The main contribution of the research:
· Developed probabilistic models for learning ambiguities using Bayesian Mixture of Experts and propagating them using discriminative density filtering.
· Generated motion capture database for controlled training and testing using a MAYA Computer Graphics model and importing various human motions to it.
· Developed several algorithms for efficient learning of low dimensional, non-linear embeddings for standard human activities.
· Developed stable multilevel 2D image encodings for semi-supervised learning of discriminative models for 3D human pose prediction.
· Investigated techniques for making discriminative 3D human pose prediction invariant to background clutter in 2D images.
Facial Features Tracking
Developed real time system for tracking facial features and recognizing facial Action Units (AU). The task entailed accurate, noise free tracking of features (eyes, eyebrows and mouth) across large head movements. The tracked 2D features were also used to drive 3D deformable face mask. Primary goals accomplished:
· Real time system (30 FPS) for tracking facial features using non-linear Active Shape Models (ASM) for generic head movement.
· Action Unit recognition based on eyebrows displacement and mouth deformation. 95% accurate head nod and head shake recognition.
· Accurate 3D head pose prediction using learned mixture of linear experts with 2D Shapes as predictor.
· 99.3% accurate eye blink detection based on histogram template matching.
Human Activity/Facial Expressions Recognition
Investigated the use of discriminative models for human activity recognition and facial expressions analysis. Discriminative models like Conditional Random Fields support long term contextual dependencies in the observation sequence making them robust to both broader class of activities and finely discriminating motion styles:
· Developed and compared algorithms for recognizing human motions in monocular video sequences, using models based on discriminative Conditional Random Fields (CRF), Maximum Entropy Markov Model (MEMM) and generative Hidden Markov Model (HMM).
· Developed algorithms for recognizing facial expressions by tracking facial feature shapes using discriminative models based on Conditional Random Field. Localized changes of facial features were tracked using Non-negative Matrix Factorization (NMF) shape basis.
PROFESSIONAL EXPERIENCE
Assoc. Member of Tech. Staff Mentor Graphics (India)R&D Labs June 2003 – Jan 2004
Developed printing framework for the mentor graphics tools (front end and backend EDA Tools). Involved in high level analysis and design of printing framework. The framework provided APIs to develop interface for specifying electronic design, printing to the printer, generating a preview or converting it to the PDF document. Responsibilities executed:
· Analyzed printing process on UNIX platform and programmed driver routines to communicate with the serial printer.
· Programmed routines to communicate with RIP (Raster Image Processor) for bidirectional PS and HP network printers using PJL and PCL.
· Installed and studied CUPS (Common Unix Printing Solution), LPD (Line Printer Daemon) and LPR as a part of project feasibility analysis.
· Designed APIs for the Graphics Kernel System based on Windows GDI+ and WxWindows GUI framework
· Designed GUI for printing framework using windows MFC programming and developed its functional specification.
Technologies used: VC++(6.0), MFC, Socket programming (UNIX), PS2, HPGL/2, PCL, PJL, Rational Clearcase (IBM).
PUBLICATIONS
Journals:
1. BM3E : Discriminative Density Propagation for Visual Tracking, C. Sminchisescu, Atul Kanaujia, Dimitris Metaxas, IEEE Transactions on Pattern Analysis and Machine Intelligence, 2007
2. Conditional models for contextual human motion recognition, C. Sminchisescu, Atul Kanaujia, Dimitris Metaxas, Computer Vision and Image Understanding, 2006,
Conferences:
1. Semi-supervised Hierarchical Models for 3D Human Pose Reconstruction, Atul Kanaujia, Cristian Sminchisescu, Dimitris N. Metaxas, CVPR 2007(Oral)
2. Large Scale Learning of Active Shape Models, Atul Kanaujia and Dimitris Metaxas, ICIP 2007
3. Learning Joint Top-Down and Bottom-up Processes for 3D Visual Inference, Cristian Sminchisescu, Atul Kanaujia, Dimitris N. Metaxas, CVPR 2006(Oral)
4. Recognizing Facial Expressions by Tracking Feature Shapes, Atul Kanaujia, Dimitris N. Metaxas, ICPR 2006(Oral)
5. Learning Ambiguities Using Bayesian Mixture of Experts, Atul Kanaujia, Dimitris Metaxas, ICTAI 2006
6. Learning Multi-category Classification in Bayesian Framework, Atul Kanaujia, Dimitris N. Metaxas, ACCV 2006
7. Tracking Facial Features Using Mixture of Point Distribution Models, Atul Kanaujia, Yuchi Huang, Dimitris Metaxas, ICVGIP 2006
8. Discriminative Density Propagation for 3D Human Motion Estimation, Cristian Sminchisescu, Atul Kanaujia, Zhiguo Li, Dimitris N. Metaxas, CVPR 2005(Oral)
9. Conditional Random Fields for Contextual Human Motion Recognition, Cristian Sminchisescu, Atul Kanaujia, Zhiguo Li, Dimitris N. Metaxas, ICCV 2005
10. Conditional Visual Tracking in Kernel Space, Cristian Sminchisescu, Atul Kanaujia, Zhiguo Li, Dimitris N. Metaxas, NIPS 2005
11. Dynamically Adaptive Tracking of Gestures and Facial Expressions, Dimitris N. Metaxas, Gabriel Tsechpenakis, Zhiguo Li, Y. Huang, Atul Kanaujia, ICCS 2006
12. Dynamic Tracking of Facial Expressions using Adaptive, Overlapping Subspaces, Dimitris N. Metaxas, Atul Kanaujia, Zhiguo Li, ICCS 2007
Workshop Papers/Technical Reports:
1. Spectral Latent Variable Models for Perceptual Inference, Atul Kanaujia, C. Sminchisescu, Dimitris Metaxas, Rutgers Univ. Tech. Report, DCS-TR-610, Feb. 2007
2. Emblem Detections by Tracking Facial Features, Atul Kanaujia, Y. Huang, Dimitris Metaxas, CVPR Workshop on semantic learning, 2006,
3. Learning to Reconstruct 3D Human Motion from Bayesian Mixture of Experts, A Probabilistic Discriminative Approach, C. Sminchisescu, Atul Kanaujia, Zhiguo Li, Dimitris Metaxas, Technical Report, University of Toronto, Oct. 2004
Patent:
"System and Method for Tracking Facial Features,", Atul Kanaujia and Dimitris Metaxas, Rutgers Docket 07-015, Provisional Patent # 60/874,451 filed December,12 2006.
GRADUATE COURSEWORK
Computer Networks, Operating System Design, Advanced Compiler Design, Database Theory, Data Structures and Algorithms, Distributed Computing, Numerical Analysis, Computer Vision, Data Mining Theory. Course projects included :
· Developed User Level thread library similar to POSIX PThreads.
· Implemented centralized and decentralized Distributed File Server
· Implemented reliable connection over User Datagram Protocol which included adding features of TCP to UDP packets
· Worked on Red Hat Linux 6.2 Kernel to modify IDE hard disk driver for file replication and file access optimization
· Worked on gene mining project to provide GUI to SQL queries using JAVA Swing and IBM Visual Age tool
· Designed and Implemented Web based Library System. CGI programming was done using JDBC and Java Servlets
TEACHING EXPERIENCE
Teaching Assistant Rutgers University Aug 2000 – May 2003
Conducted course recitations, quizzes and examination proctoring. Graded exams and assisted students with projects and assignments. Taught graduate course in Programming Languages and Compilers, Undergraduate courses in Computer architecture, Operating System Design and Programming Languages.
OTHER ACTIVITIES
· Session chair for ICTAI 2006, Washington DC.
· Program Committee for ICCV 2007, Reviewer for ICPR 2006, ACCV 2006
· External Reviewer for ICPR 2006, CVPR 2005, CVPR 2006, CVPR 2007
· Research work reported on front page of New Jersey’s largest selling daily Star Ledger.
COMPUTER SKILLS
· Programming - C, C++, VC++ 2005, Java, Java Swing, Pascal, Prolog, bash, awk, sed and perl scripts
· Operating Systems - UNIX, Red Hat Linux 5.0, X Windows
· Statistics and Vision Tools – Matlab, Netlab, OpenCV
· Database - Active Server Pages, JDBC, Java Servlets, XML, XSLT, XPath and DHTML
REFERENCES
Dr. Dimitris MetaxasDirector of CBIM Lab (Computational
Biomedicine Imaging and Modeling) Lab,
Professor at DCS, Rutgers University,
Ph: (732)-445-2914
Email: / Dr. Cristian Sminchisescu
Research Faculty at TTI-Chicago
(Toyota Technological Institute, Chicago)
Visiting Faculty at CBIM, Rutgers University
Ph.: (773)-834-2622
Email:
Dr. Vladimir Pavlovic
Assistant Professor at DCS Rutgers Univ.
312 CoRE, 110 Frelinghuysen Road
Ph: (732) 445-2654
Email: / Dr. Ahmed Elgammal
Assistant Professor at DCS Rutgers Univ.
316 CoRE, 110 Frelinghuysen Road
Ph.: (732)-445-0021
Email: