cvg









Dr. Ali Shahrokni

Research Fellow
Computational Vision Group,
School of Systems Engineering,
University of Reading
P.O. Box 225, Whiteknights,
Reading, UK, RG6 6AY. (map)
Tel: +44 118 378 7641, Fax: +44 118 975 1822.
robots



publications

personal
I joined the Computational Vision Group at the University of Reading in October 2008. I am mainly interested in research and  industrial applications of camera calibration, tracking and 3D reconstruction and rendering using arbitrary (online) images. I am currently part of the EU project Co-Friend supervised by Dr. James Ferryman. The aim of this project is to design a framework for understanding human activities in real environments.

Previously I was a post-doctoral research assistant at the  Active Vision Group in University of Oxford where I was doing research on  temporal constraints on image based rendering of video sequences as well as 3D reconstruction and rendering applied to Visual portrayal using a Visual Query.

I received my PhD in Computer Vision in 2005 from the Computer Vision Lab, Ecole Polytechnique Fédérale de Lausanne (EPFL) where I was supervised by professor Pascal Fua. My Ph.D. thesis is entitled: "Probabilistic modeling of texture transition for fast tracking and delineation" [read more].









News

  • PETS 2009 was successfully held in Miami on 25 June 2009.  A follow-on  Winter PETS workshop, using the PETS09 datasets, will take place in Snowbird, Utah, on December 9th 2009, as part of the IEEE Winter Meetings. More details including Call for Papers here.
  • The new PETS 2009 crowd analysis dataset is now available here.  This dataset contains a unique collection of different crowd movements in an outdoor environment observed by up
    to 8 synchronised cameras.  
  • I am co-organising the 11th IEEE international workshop on Perfomance Evaluation of Tracking and Surveillance, PETS 2009, in association with CVPR 2009. See the PETS2009  website for more information and call for papers.

  • Check here for a non-exhaustive selection of interesting papers from ECCV 2008, Marseille, France.






Research Interests






Video Synthesis Using Online Photos


A visual query is a  small image segment of a landmark as shown here. Current state of the art visual search engines such as  [Philbin et al’07] allows automatic retrieval of images of a landmark using the given query. We use the retrieved results to build a system that is capable of automatic scene visualisation (portrayal) from the visual query and online photos. [see project page]

 
query arrow rad arrow scene


sighs




Novel View Synthesis

Temporal consistency is a desirable feature for image based rendering of video sequences. However, it remains a challenge to deal simultaneously with depth and colour discontinuities in rendering a 3D scene correctly from a moving virtual camera. To this end, methods based on multiple depth maps and other volumetric approaches to model the 3--D scene fail to suggest tractable generative models capable of rendering high quality images.

Instead of a volumetric approach, we formulate the temporal consistency in novel view synthesis of video frames as a discrete labelling problem. The labels are the modes of a probability distribution in depth and colour space for each rendered pixel. The synthesised video sequence is then governed by an MRF over pixel colours in space-time domain.

More results can be found here.

multiple multiple multiple
Spatio-temporal MRF rendering of 3 frames reduces flicker (note inside the red circle) in the output sequence
multiple multiple multiple
Individual MRF optimisation for each output frame




dino3d




Texture

Fast texture-segmentation approach can  be used to develop robust real-time or interactive 2-D and 3-D model-based contour tracking and detection. It relies on detecting texture boundaries in the direction normal to the contour boundaries and using a Hidden Markov model to link these boundary points in the other direction. The probabilities that appear in this computation closely relate to texture entropy and Kullback-Leibler divergence [read more].
zeb




Deformable Outline Tracking

The motion of the target can be calculated by tracking its contours in a video sequence. A probabilistic representation of contours allows robust contour tracking in presence of texture and clutter. We use boosting to train a predictor of the conditional probability of texture transition, given the pixel intensities. Connected object contours are then obtained by maximising the posterior probability of object model parameters. Again a Hidden Markov Model is used to calculate the joint law of the conditional probabilities of contour points [read more].
outline
See video




3-D  Object Tracking

Line-search texture-based tracking provides a real-time 3–D pose estimation algorithm that retains the speed of those that rely solely on gradient properties along object contours but does not fail in the presence of highly textured object and clutter.
This is achieved by correctly integrating probabilities over the space of statistical texture models. This rigorous and formal statistical treatment results in good performance under demanding circumstances [read more].
ns
See video




3-D  Articulated Tracking

Optimisation of a suitable energy function on cross matched sample points and body contours allows tracking of articulated human body motion in 3-D dimensional space using monocular video sequences [read more].
augmented




Augmented Reality
jedi




Publications

Refereed conference papers

  • J. Berclaz, A. Shahrokni, F. Fleuret, James Ferryman and P. Fua. Evaluation of Probabilistic Occupancy Map People Detection for Surveillance Systems. Proceedings of IEEE International Workshop on Performance Evaluation of Tracking and Surveillance (PETS 2009). Miami, USA, 2009 [pdf]
  • J. Ferryman and A. Shahrokni. An Overview of the PETS2009 Dataset. Proceedings of IEEE International Workshop on Performance Evaluation of Tracking and Surveillance (PETS 2009). Miami, USA, 2009 [pdf]
  • A. Ellis, A. Shahrokni and J. Ferryman. Overall Evaluation of the PETS2009 Results. Proceedings of IEEE International Workshop on Performance Evaluation of Tracking and Surveillance (PETS 2009). Miami, USA, 2009 [pdf]
  • A. Shahrokni, P.H.S. Torr. I. Reid. From Visual Query to Visual Portrayal. Proceedings of British Machine Vision Conference, 2008 [ pdf | BibTeX]
  • A. Shahrokni, O. Woodford. I. Reid. Temporal Priors for Novel Video Synthesis. In proceedings of Asian Conference on Computer Vision, Tokyo Japan, 2007 [ pdf | BibTeX ]
  • A. Shahrokni, T. Drummond, P. Fua. Fast Texture-Based Tracking and Delineation Using Texture Entropy. In proceedings of International Conference on Computer Vision, Beijing, China, 2005 [ pdf | BibTeX]
  • A. Shahrokni, F. Fleuret, P. Fua. Classifier-based Contour Tracking for Rigid and Deformable Objects. In proceedings of British Machine Vision Conference, Oxford, England 2005 [ pdf | BibTeX]
  • A. Shahrokni, V. Lepetit, T. Drummond, P. Fua. Markov-based Silhouette Extraction for Three Dimensional Monocular Body Tracking in Presence of Cluttered Background. In proceedings of British Machine Vision Conference, Kingston, England 2004 [ pdf | BibTeX]
  • A. Shahrokni, T. Drummond, P. Fua. Texture Boundary Detection for Real-Time Tracking. In proceedings of European Conference on Computer Vision, Prague, Czech Republic, May 2004 [ pdf | BibTeX]
  • V. Lepetit, A. Shahrokni, P. Fua. Robust Data Association. In proceedings of International Conference on Computer Vision and Pattern Recognition, USA, June 2003 [ pdf | BibTeX]
  • A. Shahrokni, V. Lepetit, and P. Fua. Bundle Adjustment for Markerless Body Tracking in Monocular Video Sequences. In ISPRS workshop on Visualization and Animation of Reality-based 3D Models, Switzerland, 2003 [ pdf | BibTeX]
  • A. Shahrokni, L. Vacchetti, V. Lepetit, P. Fua. Polyhedral Object Detection and Pose Estimation for Augmented Reality Applications. In proceedings of Computer Animation 2002, Geneva, Switzerland [ pdf | BibTeX]
  • H. Soltanian-Zadeh, A. Shahrokni, R. Zoroofi. A Voxel-Coding Method for Quantification of Vascular
    Structure from 3D Images. In proceedings of SPIE Medical Imaging Conference, San. Diego, CA, Feb.
    17-23, 2001 [ pdf | BibTeX]
  • A. Shahrokni, R. Zoroofi, H. Soltanian-Zadeh. A Fast Skeletonization Algorithm for 3-D Elongated Objects. In proceedings of SPIE Medical Imaging Conference, San. Diego, CA, Feb. 17-23, 2001[ pdf | BibTeX]
  • A. Behrad, A. Shahrokni, S. A. Motamedi and K. Madani. A Robust Vision-Based Moving Target Detection and Tracking System. In proceedings of Image and Vision Computing conference (IVCNZ2001), University of Otago, Dunedin, New Zealand Nov. 26-28, 2001
  • A. Shahrokni. Fuzzy-Based Segmentation of 2- and 3-D images. In proceedings of the 3rd symposium
    on intelligent systems, University of Tehran, Tehran, Iran, April 2000

Journal Papers

  • A. Shahrokni, T. Drummond, F. Fleuret, P. Fua. Classification-based Probabilistic Modeling of Texture Transition for Fast Line Search Tracking and Delineation. Accepted for publication IEEE Transactions on Pattern Analysis and Machine Intelligence, 570-576, 2009 [ pdf | BibTeX]
  • Ali Shahrokni, Oliver Woodford, Ian Reid. Temporally-coherent Novel Video Synthesis Using Texture-based Priors. Accepted for publication in IPSJ Transactions on Computer Vision and Applications, 2008 [pdf]
  • M. Maddah, H. Soltanian-Zadeh, A. Afzali-Kusha, A. Shahrokni and Z. Zhang. Three-dimensional analysis of complex branching vessels in confocal microscopy images. Computerized Medical Imaging
    and Graphics, Volume 29, Issue 6, Pages 487-498, September 2005 [ pdf ]
  • H. Soltanian-Zadeh, A. Shahrokni, M. Khalighi, Z. Zhang, R. Zoroofi, M. Maddah and M. Chopp. 3-D quantification and visualization of vascular structures from confocal microscopic images using skeletonization and voxel-coding. Computers in Biology and Medicine, Volume 35, Issue 9, Pages 791-813, November 2005 [ pdf ]










Dr. Ali Shahrokni, email: a.shahrokni AT reading.ac.uk