Instituto de Sistemas e Robótica

Joćo Paulo Costeira
Associate Professor
Department of Electrical and Computer Eng.
Instituto Superior Técnico
Email- jpc AT
Phone (Office) +351-218418292, Ext 2292

Postal Address - Torre Norte, Av. Rovisco Pais, 1049-001  Lisbon, Portugal


PhD in Electrical and Computer Eng. - IST (1995), PhD Thesis -  Motion Segmentation(PDF) -Advisors: Takeo Kanade and Joćo Sentieiro

Visiting Scientist  - Robotics Institute - Carnegie Mellon University (1992-1995) -(research report )

Researcher@ Signal and Image Processing Group  
Coordinator of Thematic Area   - ISR-Associated Laboratory (2007-2011)
Group Coordinator - Signal and Image Processing Group SIPg

(2007-2018) Co-director of the Dual PhD program in ECE and Robotics Carnegie Mellon|Portugal

(2014-2018) Scientific Director Carnegie Mellon|Portugal

See My Google Scholar Profile or  here
Scientific Interests

Computer Vision, 3D Reconstruction, Structure from Motion, Recognition

Computer Vision and Image Processing (PIV -2011 )

Past courses :  Signals&Systems, Circuit Analysis, Control
PhD Students
Joao Maciel  (2001), Optimal Solution to Correspondence Problems

Ricardo Oliveira (2006) Optimal Correspondence through the use of rank constraints

Nuno Pinho da Silva (2010)  Robust Nonmetric Perception of Moving Rigid Bodies

Ricardo Ferreira (2010) Reconstruction of isometrically embedded flat surfaces from scaled orthographic image data co-advised  with  J. Xavier

Manuel Marques  (2011)   "The Role of Rigidity in Structure from Motion: Unique Solutions in 3D Reconstruction and Object Matching"

Susana Brandao(2015)Heat-Based Descriptors for 3D Multiple View Object Recognition - co-advised with Manuela Veloso in  Carnegie Mellon|Portugal

Ricardo Cabral  (2015)Unifying Low Rank Models for Visual Learning - co-advise with Fernando De la Torre & A. Bernardino in Carnegie Mellon|Portugal

Jayakorn Vongkulbhisal  (2018)Discriminative Optimization: Theory and Applications to Computer Vision - co-advise with Fernando De la Torre in Carnegie Mellon|Portugal

Qiwei Han, Essays on Consumer Switching and Search Behavior co-advised with Pedro Ferreira(CMU) in Carnegie Mellon Portugal

Shanghang Zang , co-advised with José Moura (CMU) in Carnegie Mellon Portugal
Current students
Joćo Carvalho, co-advised with Manuel Marques in NETSyS program

Beatriz Quintino, co-advised with Joćo Pedro Gomes in NETSyS program

Jude Mukundane, co-advised with Chris Csikszentmihallyi in NETSyS program
A Summary of recent works (keynote address at the "International Conference on Transport Survey Methods"
My (Computer) Vision
I've been pursuing the idea of creating an artificial vision "machine"  since 1986. Check  CMU's Virtualized Reality  if you are not from the field. I've  had a lot of fun on many subjects but some problems never left my mind: 3D reconstruction and "the correspondence problem".  Those are the problems I've shared with my PhD students and people I had the previlege to collaborate with.
Current Projects

Vehicle Counting in City Scale  Camera  Networks

 3D Point Cloud Registration for Robotic Applications Discriminative Optimzation


People counting in public infrastructures
3D Reconstruction 

Multiple motion segmentation for rigid and articulated moving objects By tracking points in a video sequence with several objects moving, points are grouped according to "rigidity".
Joao P. Costeira, Takeo Kanade - A Multibody Factorization Method for Independent Moving Objects,  International Journal on Computer Vision, 29(3), pp.159-179, Kluwer Academic Publishers, 1998
Nuno P da Silva Costeira, J. P.. " The Normalized Subspace Inclusion: Robust Clustering of Motion Subspaces", The Twelfth IEEE International Conference on Conputer Vision (ICCV), Kyoto, Japan, September, 2009.
Nuno P da Silva  Costeira, J. P.. " Subspace segmentation with outliers: a Grassmannian approach to the maximum consensus subspace", IEEE  Conference on Conputer Vision and Pattern Recognition (CVPR), Anchorage, Alaska, USA, June, 2008.

Thanks to Vargas for the rendering
Factorization with missing and degenerate data

The 3D model of "Casa da Musica"  is computed from point coordinates in the images.

Main contributions:
- Uncalibrated reconstruction (images obtained  on google images).
- Very dissimilar images
- Visible are can be degenrate (planar).
State of the art brakes down.

This method was used for face recognition .
Marques, M., and Costeira, J.. "Estimating 3D shape from degenerate sequences with missing data", Computer Vision and Image Understanding, 113(2):261-272, February 2009.  
Marques, M., and Costeira, J.. "3D face recognition from multiple images: a shape-from-motion approach", FG2008 - 8th IEEE International Conference on Automatic Face and Gesture Recognition, Amsterdam, The Netherlands, September 2008. 
Marques, M., and Costeira, J.. "Optimal shape from motion estimation with missing and degenerate data", WACV-Motion 2008 - IEEE Workshop on Application of Computer Vision, Copper Mountain, CO, USA, January 2008.

Reconstruction of "paper like" surfaces. The flat surface is reconstructed from point tracks (with missing data). 
 Ricardo FerreiraJoćo XavierJoćo Paulo Costeira, Shape from motion of nonrigid objects: the case of isometrically deformable flat surfaces, Proc. of BMVC 2009 - British Machine Vision Conference, London, UK, 2009

Ricardo Ferreira, Joćo Xavier, Joćo Paulo Costeira,Reconstruction of Isometrically Deformable Flat Surfaces in 3D from Multiple Camera Images,  Proc. of ICASSP 2009 - IEEE International Conference on Acoustics, Speech, and Signal Processing, Taipei, Taiwan, 2009 

Page under construction :-)