033001195_002_000013Name : Nicolas MoŽnne-Loccoz

Institution : Sagem Sťcuritť

 

Nicolas MoŽnne-Loccoz received the BA degree in computer science from JosephFourier University of Grenoble in 1999. He received the MS degree in computer science from Nice-Sophia-Antipolis University in 2001.He received the PhD degree from University of Geneva in 2005.From 2005 to 2007 he was extending his research on video event characterization as a post-doctoral in the Viper team at the University of Geneva. In 2007 he joined the biometric research unit of Sagem Sťcuritť where he works on several topics to improve face-based identity management solutions such as face detection, tracking and recognition.

 

Publications :

MoŽnne-Loccoz, N., Janvier, B., Marchand-Maillet, S., & Bruno, E. (2006). Handling Temporal Heterogeneous Data for Content-Based Management of Large Video Collections. Multimedia Tools and Applications, 31, 309-325.

 

MoŽnne-Loccoz, N., Bruno, E., & Marchand-Maillet, S. (2004). Knowledge-based Detection of Events in Video Streams from Salient Regions of Activities. Pattern Analysis and Applications (PAA), special issue Video Event Mining, 7(4), 422-429.

 

 

 

Title of Project : Capturing identities without interaction:ďFace on the flyĒ

 

The identification of people evolving within public facilities becomes more and more required to ensure its security. But the identification process should be as transparent as possible for users. The first reason is that the flow of people should not be disrupted. The second reason comes from lessons learned in the deployment of large-scale biometric systems : interaction between users and the system is error prone and should be minimized. However reducing these interactions imply alleviating the acquisition constraints in terms of user distance, pose and motion relatively to the sensors.In this work we present a face-based biometric identification system which acquires on the fly the biometric signatures of people, i.e. with a minimal amount of interaction. The solution is based on the reconstruction of a set of ideal frontal views of the user getting over the acquisition volume.First a set of multi-view face images of the user is collected by the video cameras covering the acquisition volume. Underlying algorithms ensure the robustness to the user motion and pose. However, the resulting views do not permit to achieve identification with a sufficient level of accuracy because of the face orientation unpredictability. Hence a 3D model of the face is estimated over timefrom these multi-view acquisitions and a set ofideal face biometric signatures is computed from this estimate that finally permits to achieve highly accurate identification of people.

 

Contact : nicolas.moenne-loccoz@sagem.com