SITe - Stage de recherche tracking vidéo |
 |
Face tracking/extraction from a video stream for facial recognition
A facial recognition on the move system can be divided in three modules:
- A tracking module that detects and extracts faces from the video stream.
- A encoding module that converts the face picture coming from the tracking module into a template (i.e. a unique signature as a number)
- A matching module that compares the above template to the ones stored in a watch list database. In case of match, an alert is generated.
The aim of the study proposed here is to specify and develop video algorithms to detect, to track and to optimise in the video the faces that will be sent to the encoder/matcher modules.
The study will explore different ways to build the best 2D face image of people crossing the field of view of a Mega Pixels camera:
- Detection and tracking of one or several faces
- Evaluation of the quality of the faces / frame
- Use of the different frames containing the same face in order to build a complete normalized face 2D image if possible
Normalized 2D image face can be for example (non exhaustive and to be defined precisely):
- Face will be in front view
- Horizontal eyes exposure
- At least 40 pixels between the eyes
- At least 64 grey level on the face
- No background
The study will also propose evaluations methods and results.
Keywords:
Image/ video processing, facial detection, tracking and recognition
Requirements:
knowledge in image/ video processing, C/ C++
Partner:
THALES
Contact :
Maria TROCAN
|
Recherche 
