Ankush Roy
|
Computer Science Department
University of Alberta
ATH 106, Athabasca Hall
Edmonton, AB T6G2E8
Email:
|
I am a Masters student working on computer vision and machine learning with
Prof. Martin Jagersand,
focusing on search methods in Object Tracking, in the Computer Vision and Robotics Lab, in University of Alberta.
I also collaborated with
Prof Utpal Garain of
Indian Statistical Institute and Prof Anjan Rakshit
of Jadavpur University during my undergraduate years.
News:
- Thesis Work
- January 2015: Paper on benchmarking tracking videos for manipulational tasks ("Tracking Benchmark and Evaluation for Manipulation Tasks") accepted in International Conference on Robotics and Automation ( ICRA ) 2015
- January 2015: Paper ("Machine Authentication of Paper Currency: an Experiment on Indian Banknotes") on full system detecting counterfeit currency notes in Indian economy accepted (with changes) in International Journal on Document Analysis and Recognition (IJDAR)
- September 2014: Dataset of videos to evaluate registration based trackers is made available. An initial example of three trackers are evaluated and results
are posted online.
- June 2013: Our paper "A Probabilistic Model for Reconstruction of Torn Forensic Documents " describing a model to reconstruct torn natural images, will appear in ICDAR2013.
Projects:
- Gradient based search methods in tracking assume locally small motion of the object which
make them susceptible to track objects moving fast. Approimate Nearest Neighbour
can be used to overcome this problem as shown by
Travis et. al . I propose a
graph based approximate nearest neighbour algorithm that takes into account the
sequential data assumption which is explicit in video data to make tracking both faster
and more robust. We show comparitive results with many state of the art registration trackers.
Technical Report: Graph based Nearest Neighbour in Registration based Object Tracking 
Ankush Roy, Xi Zhang and
Martin Jagersand
- Dataset to evaluate registration based trackers,
This dataset contain 100 video sequences thoughtfully recorded to evaluate registration based trackers. We also provide an
initial evaluation of some of the existing trackers in literature. A new evaluation metric is also proposed to judge a tracker's
robustness to high inter frame motion.
Technical Report: Tracking Benchmark and Evaluation for Manipulation Tasks
Ankush Roy, Xi Zhang, Nina Wolleb, Camilo Perez and
Martin Jagersand
- Detection based initialisation of
8 d.o.f. trackers. This is handy if we have the model of the objects and a robot is teleoperated by just selecting
the category of the object. This also augments the tracker's performance by re - initialising the tracker when it drifts.
- Multiple Kernel Learning We study
the problem of multiple kernel learning (MKL) in a classification setting. We first examine the kernel alignment
metric and show that maximizing the alignment of a kernel with the target kernel YYT corresponds to a constrained
minimization of the margin loss of a weighted Nearest-Neighbor (NN) classifier.
Publications:
International Conference on Robotics and Automation, 2015
International Conference on Document Analysis and Recognition, 2013.
21st International Conference on Pattern Recognition, 2012.
National Conference on Computer Vision, Pattern Recognition, Image Processing and Graphics (NCVPRIPG), 2011.
National Conference on Emerging Trends and Applications in Computer Science (NCETACS) , 2011.
- Won the best paper of the conference
Indian Conference on Computer Vision, Graphics and Image Processing (ICVGIP), 2010.
-
Technical Report: Machine Assisted Authentication of Paper Currency: an Experiment on Indian Banknotes
Ankush Roy, Biswajit Halder
Utpal Garain and
David Doermann.
International Journal of Document Analysis and Recognition (IJDAR) , 2015. (Accepted: final copy due)
[GOOGLE SCHOLAR]
[RESUME]
Codes
Teaching
- CMPUT 415: Compiler Design, with Dr. Paul Berube in Fall 2012
- Graded 4 Assignments
- Evaluated Final Project
- CMPUT 206: Introduction to Image Processing, with Prof Nilanjan Ray in Winter 2013
- Presented Lab Tutorials
- Graded 2 Assignments
- CMPUT 379: Operating System and Concepts, with Bob Beck in Fall 2013
- Graded Lab Assignments
- Developed and Presented Lab tutorials
- CMPUT 379: Operating System and Concepts, with Martha White in Winter 2014
- Graded Lab Assignments
- Developed and Presented Lab tutorials
- Guest lecturer for sever design
- CMPUT 379: Operating System and Concepts, with Mohammad Bhuiyan in Fall 2014
- Graded Lab Assignments
- Developed and Presented Lab tutorials
- Course Material: Tutorials and Codes
Other Interests
- Soccer: I have been playing since the age of 12. I played in the school team and in college.
- Table Tennis
- Squash
- I like to read about how programming languages are designed.
