Service interruption on Monday 11 July from 12:30 to 13:00: all the sites of the CCSD (HAL, EpiSciences, SciencesConf, AureHAL) will be inaccessible (network hardware connection).
Skip to Main content Skip to Navigation
Theses

Reconnaissance de l’orientation des piétons par les réseaux de Capsules dans un environnement non structuré pour un ADAS

Abstract : Road accidents are the first cause of death for those who are under 30 years old. Represented as the most vulnerable road user, the pedestrian constitutes 23% of all road fatalities. This thesis is part of the research conducted on the application of deep learning methods for pedestrian safety. In this work, we propose a pedestrian orientation detection system that could be integrated into Advanced Driver Assistance Systems (ADAS) to alert the driver of the presence of a pedestrian. To this end, we created a new pedestrian orientation database called "SAFEROAD Dataset" recorded from different Moroccan cities using a monocular camera in a moving vehicle. This database contains 8894 images of pedestrians that are manually annotated in 4 and 8 directions. We then proposed a new technique to detect pedestrian orientation using Capsule networks. The training and evaluation of this technique is done on our SafeRoad, Daimler and TUD base. The algorithm trained on the SafeRoad database is then applied for pedestrian orientation recognition on video sequences taken from the JAAD dataset. Finally, we propose in this thesis a pedestrian-vehicle accident prevention system, which integrates the pedestrian’s orientation for the evaluation of the risk of an accident. This system takes into consideration the presence of undisciplined pedestrians and unstructured roads.
Complete list of metadata

https://tel.archives-ouvertes.fr/tel-03630693
Contributor : ABES STAR :  Contact
Submitted on : Tuesday, April 5, 2022 - 11:09:38 AM
Last modification on : Wednesday, April 6, 2022 - 3:37:04 AM
Long-term archiving on: : Wednesday, July 6, 2022 - 6:42:00 PM

File

DAFRALLAH.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-03630693, version 1

Citation

Safaâ Dafrallah. Reconnaissance de l’orientation des piétons par les réseaux de Capsules dans un environnement non structuré pour un ADAS. Réseau de neurones [cs.NE]. Normandie Université; Université Ibn Tofail. Faculté des sciences de Kénitra, 2021. Français. ⟨NNT : 2021NORMIR29⟩. ⟨tel-03630693⟩

Share

Metrics

Record views

49

Files downloads

8