dspss.vef.gov

[Kay]

Hi all,

Please find below an announcement for a PhD opening at Ecole des Mines de Paris.

========================================

PhD Proposal

Features and machine-learning algorithms, using motion information, for multiclass object or action recognition"

Location: Ecole des Mines de Paris
Robotics Centre (CAOR) - http://caor.ensmp.fr
60 boulevard Saint-Michel, 75272 Paris Cedex 06
PhD Supervisor: Bogdan Stanciulescu
Starting date: September 2008, 3 years
Financing: fellowship “Ecole des Mines” or “Armines”
Key-words: machine learning, image processing, pattern recognition, tracking, intelligent transportation systems, video-surveillance.

Context
This PhD thesis is proposed within the framework of the Robotics Centre of the Ecole des Mines de Paris. The Robotics Centre is conducting applied research on the following fields: intelligent transportation systems (ITS), virtual and augmented reality, 3D environment reconstruction, advanced control systems, video-surveillance. The Robotics Centre is participating in R&D projects with French and European industry and academic institutes. Successful applications have been developed and transferred to the industry and lead to the creation of several start-up companies. Among them we quote the RTMaps prototyping system, used for a real-time environment analysis and multi-sensor fusion in various applications, such as pedestrian and car detection, advanced driving assistance, etc.
Ecole des Mines is a prestigious French engineering school located in the latin quarter of Paris.
Subject description
This PhD proposal is focused on multi-class object (i.e. people, vehicles, faces, etc) and action detection and recognition, in dynamical environments, using mono-camera image sequences. To this aim, research on the following axis is to be considered:
• Machine learning algorithms with applications in computer vision;
• Spatio-temporal features for object and class of objects (people in particular).
The Robotics Centre of Ecole des Mines has developed specific tool and algorithms for classifier learning, using object cropped-images from image sequences. This tool is based on the AdaBoost algorithm, which is gathering and weighting several weak classifiers, each one representing an object feature.
One goal of this thesis proposal is to extend the existing tools and framework to multi-class machine learning algorithms (AdaBoost, SVM, others), ensuring the simultaneous classification and recognition of several object categories. The results of this research results could be used for semantic scene analysis, in video-surveillance and ITS applications.
The other goal of this proposal is to construct a new class of spatio-temporal visual features, integrated within the same research framework. These features will
integrate the motion (temporal) and the spatial information, in order to simultaneously quantify the shape and the object (or their components) dynamics.
Among the possible applications of the student’s research, the object’s actions recognition and action-type classification (lateral movement, static pose, etc.) could be explored. A way in achieving this could be a statistical learning approach on the family of spatio-temporal features.
The candidate is requested to follow scientific training courses required by Ecole des Mines de Paris. He/She shall integrate a dynamic research team; thus he must show his ability to work in team and participate to the scientific activities of the laboratory.

Candidate Profile
The successful candidate must hold a Master degree and have the following skills:
• Good knowledge of machine learning algorithms;
• Good level in image processing and computer vision.
• Fluent in English and good level in French.
• C/C++ or Java programming language.
• Taste for experimental validation.
• Driving licence is appreciated.

Contact
Scientific Contact: Bogdan Stanciulescu : E-mail : Bogdan.Stanciulescu@ensmp.fr
Tél : +33 – (0)1.40.51.94.98
Application Contact : Christine Vignaud : E-Mail: Christine.Vignaud@ensmp.fr,
Tél :+33 – (0)1.40.51.92.55