Ifsttar PhD subject

French version
Detailed form :

Title : Visibility restoration in images and videos acquired in adverse weather conditions

Main host Laboratory - Referent Advisor   -

Director of the main host Laboratory   -

PhD Speciality Vision par ordinateur-Traitement d'images

Axis of the performance contract 1 - COP2017 - Efficient transport and safe travel

Main location Marne-la-Vallée

Doctoral affiliation UNIVERSITE PARIS-EST

PhD school MATHEMATIQUES ET SCIENCES ET TECHNOLOGIES DE L'INFORMATION ET DE LA COMMUNICATION (MSTIC)

Planned PhD supervisor BREMOND Roland  -  Université Gustave Eiffel  -  COSYS - PICS-L

Planned financing Contrat doctoral  - Ifsttar

Abstract

Some weather conditions such as rain and fog contribute to reduced visibility, which can cause traffic accidents. Using advanced driver assistance systems (ADAS), the risk of accidents is reduced due to a large number of on-board sensors equipped with specific processors and software. However, these systems are not always efficient enough to deal effectively with degraded atmospheric conditions. Then, it becomes necessary to integrate sensors and image processing in order to attenuate the visual effects caused by degraded conditions. The use of image and video restoration algorithms makes it possible to attenuate these effects.
The aim of this thesis is to design algorithms running as close as possible to real time to improve the visibility of the scene by restoring the images acquired in degraded weather conditions. We worked on fine particles including mist, fog and dust. Two approaches are compared : a physical-based approach involving a physical model of fog, and a learning-based approach. Several algorithms have been proposed to address these issues including a single image fog removal algorithm based on physical prior, and a video fog removal algorithm with a deep learning technique. Moreover, the lack of video datasets containing fog encouraged us to create one. Evaluations were conducted for each of the algorithms in order to assess their effectiveness.

Keywords : Images,Restoration,Adverse weather conditions

Main host Laboratory - Referent Advisor	-
Director of the main host Laboratory	-
PhD Speciality	Vision par ordinateur-Traitement d'images
Axis of the performance contract	1 - COP2017 - Efficient transport and safe travel
Main location	Marne-la-Vallée
Doctoral affiliation	UNIVERSITE PARIS-EST
PhD school	MATHEMATIQUES ET SCIENCES ET TECHNOLOGIES DE L'INFORMATION ET DE LA COMMUNICATION (MSTIC)
Planned PhD supervisor	BREMOND Roland - Université Gustave Eiffel - COSYS - PICS-L
Planned financing	Contrat doctoral - Ifsttar
Abstract Some weather conditions such as rain and fog contribute to reduced visibility, which can cause traffic accidents. Using advanced driver assistance systems (ADAS), the risk of accidents is reduced due to a large number of on-board sensors equipped with specific processors and software. However, these systems are not always efficient enough to deal effectively with degraded atmospheric conditions. Then, it becomes necessary to integrate sensors and image processing in order to attenuate the visual effects caused by degraded conditions. The use of image and video restoration algorithms makes it possible to attenuate these effects. The aim of this thesis is to design algorithms running as close as possible to real time to improve the visibility of the scene by restoring the images acquired in degraded weather conditions. We worked on fine particles including mist, fog and dust. Two approaches are compared : a physical-based approach involving a physical model of fog, and a learning-based approach. Several algorithms have been proposed to address these issues including a single image fog removal algorithm based on physical prior, and a video fog removal algorithm with a deep learning technique. Moreover, the lack of video datasets containing fog encouraged us to create one. Evaluations were conducted for each of the algorithms in order to assess their effectiveness.
Keywords :	Images,Restoration,Adverse weather conditions

List of topics

Applications closed