Research


I am currently carrying out a joint PhD at 
ViCOROB (Universitat de Girona) and LE2I (Université de Bourgogne). My researches are focused on developing a Computer Aided Diagnosis (CAD) system for prostatic biopsy guidance and follow-up using multi-modal medical imaging.


Traffic Signs Detection - Recognition - Tracking







We are currently working on a traffic signs detection-recognition-tracking system.

This work is done in collaboration with LE2I (France - Le Creusot) under supervision of Yohan Fougerolle.

Our contribution is threefold:
  • Proposition of a new naive detection based on color and shape retrieval.
  • Using the information of the detection, we propose to recognize traffic sign using machine learning methods.
  • In order to perform to accelerate the process of detection-recognition, we propose to introduce a tracking module information already extracted in the two previous stages.

The source code of this implementation is available in GitHub.

 


Real-Time Visual Object Tracking

We develop a method allowing to track any kind of object inside a video.

Our method belongs to the group of features-based tracking method. In fact, features are detected in two consecutive images and correspondences are found.

Geometric transformation is estimated using the previous matching by the aid of a robust estimator.

The method proposes the following advantages:
  • Real-time computation.
  • Scale invariant.
  • Rotation invariant.
  • Partially robust to illumination changes.
  • Partially robust to occlusions.
More details can be found in the above publication:


G. Lemaitre, E. Vargiu, J.A. Lorenzo Fernández and F. Miralles, 
"Real-Time 2D Face Detection and Features-based Tracking in Video",
IADIS Multi Conference in Computer Science in Computer Graphics, Visualization, Computer Vision and Image Processing 2012. Lisbon: Portugal (July 2012)


 
 

Pruning AdaBoost for Continuous Sensors Mining Applications


In this paper, pruning techniques for the AdaBoost classifier are evaluated specially aimed for continuous learning in sensor mining applications. To assess the methods, three pruning schemes are evaluated using standard machine-learning benchmark datasets, simulated drifting datasets and real world cases.

Early results obtained show that the pruning methodologies approach and sometimes out-perform the no-pruned version of the classifier, being at the same time more easily adaptable to the drift in the training distribution.

Future works are planned in order to evaluate the approach in terms of time efficiency and big-data extensions.

More details can be found in the above publication:

M. Rastgoo, G. Lemaitre, X. Rafael, F. Miralles and P. Casale, 
"Pruning AdaBoost for Continuous Sensors Mining Applications",
Ubiquitous Data Mining Workshop, 20th European Conference in Artificial Intelligence 2012. Montpellier: France (August 2012)