Les offres de “SAFRAN”

Expire bientôt SAFRAN

Stagiaire en Deep Learning pour la navigation de vehicule H/F

  • Stage
  • France
  • Design / UX / UI

Description de l'offre

Entité de rattachement

Safran est un groupe international de haute technologie, équipementier de premier rang dans les domaines de l'Aéronautique, de l'Espace et de la Défense. Implanté sur tous les continents, le Groupe emploie près de 58 000 collaborateurs pour un chiffre d'affaires de 15,8 milliards d'euros en 2016. Safran est une société cotée sur Euronext Paris et fait partie des indices CAC 40 et Euro Stoxx 50.

Safran est classé dans le Top 100 Global Innovators de Thomson Reuters. Safran est également classé en tête du palmarès « Happy at work », classement réalisé par le site meilleures-entreprises.com, sur le podium des entreprises préférées des jeunes ingénieurs* et dans le classement LinkedIn des entreprises les plus attractives en France.
*enquêtes Universum et Trendence  

Description du poste

Filière principale / Métier principal

Recherche, conception et développement - Mathématiques et Algorithmes

Intitulé du poste

Stagiaire en Deep Learning pour la navigation de vehicule H/F

Type contrat

Stage

Durée du contrat

6 mois

Statut

Etudiant

Temps de travail

Temps complet

Description de la mission

Context
Navigation is an essential task for autonomous vehicles and consists in monitoring and estimating the vehicle's own location and movement. Mature technologies for this purpose resort to various kinds of measurements, typically merged using methods from the field of automatic control (Kalman filters, sliding-window optimization; nonlinear observers, etc.). They are well suited to sensors returning kinematic quantities (accelerometers, gyroscopes, GPS receivers), but less to embedded cameras where images taken by the cameras are rather indirectly related to the motion of the carrier vehicle. This makes the integration of video sequences for navigation more difficult and requires an additional image processing layer for the extraction of useful information for movement estimation. To this end, various approaches have been proposed in the past: feature points tracking, optical flow computation, direct luminosity matching.
In the meantime, a new research avenue has been opened by the recent progress made by Convolutional Neural Networks (CNNs). CNNs have improved significantly performances in various traditional computer vision tasks, such as image classification, object detection, face recognition, etc., all of which were relying previously on classic image processing techniques. Such techniques are still currently in use in the visual navigation community. Most recent variants of CNN architectures and intelligent methods for generating and leveraging synthetic training data have advanced the State-of-the-Art further in other computer vision tasks such as estimation of optical flow , estimation of the transformation between pairs of images, detection of interest points [6], etc. Most of the newly addressed task are in fact composing blocks of the visual navigation pipeline. However, visual navigation itself remains a complex and difficult task and no end-to-end effective strategy has been proposed for it yet.
Properly interfacing automatic control and deep learning is the key to many current challenges of autonomous systems, and some breakthroughs can be expected on this topic in a near future. In this context, Safran proposes internships giving student the opportunity to contribute to this research field.
Internship topic
The objective of this internship is to design a Deep Neural Network that would allow a vehicle or drone to estimate its motion. The internship will be carried in two main stages:
 The intern will first develop a neural network, using concepts from Domain Adapdation litterature, for estimating the shifts between two images depicting similar scenes from different points of view.
Then, the intern will think of a manner to integrate the temporal dimension in the models designed above. The benefit is twofold: it would allow improving the accuracy and stability of the estimations and it would make possible the integration of data coming from an Inertial Measurement Unit.

Profil recherché

Profil candidat

Aptitudes et expériences souhaitées :
Formation : 5ème année d'école d'ingénieur, master recherche ou stage de césure
Langues : Anglais pour lecture scientifique
Spécialités : Mathématiques appliquées, machine learning, traitement d'image, vision par ordinateur, robotique
Qualités requises : Initiative, autonomie, capacité d'adaptation et esprit de synthèse. Goût pour la recherche
Langages de programmation : Python

Faire de chaque avenir une réussite.
  • Annuaire emplois
  • Annuaire entreprises
  • Événements