The subject is part of SAFER-LC project (H2020), starting from May 2017 for a duration of 36 months.
The main objective of SAFER-LC is to improve safety and minimize risks at and around level crossings
(LCs) by developing a fully integrated cross-modal set of innovative solutions and tools for the
proactive management and new design of level-crossing infrastructure. The project will focus both on
technical solutions, such as smart detection services and advanced infrastructure-to-vehicle
communication systems and on human processes to adapt infrastructure design to end-users and to
enhance coordination and cooperation between different stakeholders from different transportation
The objective of the work related to the postdoc subject is to identify and understand the dynamics of
the development of hazardous situations in LC environments. Oppositely to manual diagnosis,
performed by human investigations and observations, the task proposes to extend the temporal
dimension of this diagnosis by exploiting automatically big databases provided by video system
installed in LC for acquiring video sequences over long periods (several weeks). The proposed off-line
automatic video analysis system will allow to refine/correlate human diagnosis with quantitative
information (statistics and classification) extracted from the automatic system, and to extract
behavioural models of user-to-user and user-to-infrastructure (LC) interactions.
Although this problem is widely addressed by researchers in different contexts, it constitutes a niche of open questions with scientific and technological locks. Indeed, there is no generic user behavioural model for a given context, since it is necessary to take into account specificities (physic and socio-demographic) of each environment of the context, including the variety of users. Furthermore, user behaviour depends
generally to interaction that makes running one or several instantly situations, different types of users,
and the infrastructure and its topology (user-to-user and user-to-infrastructure interactions).
Hence, the objective is to develop an intelligent video surveillance system able to model and analyze
user behaviours, and extract useful information that could help understanding of the hazard. This task
proposes to exploit big data provided by video systems installed in LC environments for acquiring video
sequences over long periods (several weeks) of time, as well as by other environmental sensing
systems. The behaviour analysis will be performed off-line by video image sequence processing.
The task will be developed in collaboration with CEREMA (Toulouse, France), DLR (Germany) and NTNU
(Norway). The information that will be extracted by the surveillance system includes type and
trajectory of users, structure and layout of the infrastructure (LC). Output from other work packages
of the project will be exploited in developing the intelligent video surveillance system. For example,
typical factors behind common LC accidents will be used to select the types of LC to be evaluated in
this task; and societal dimension of human behavior should be taken into account to complete and
refine mathematical model and physical models).
Requirements: Candidate with PhD degree in computer vision/artificial intelligence. High experience
and background in video processing, machine learning and programming is searched. Excellent English
communication capabilities are required.
Deadline to apply: 30 June 2017
Information contact: Yassine RUICHEK (firstname.lastname@example.org)
|Free forum by Nabble||Edit this page|