Obaida ALRAZOUK defends his doctoral thesis on thirsday, December 9th, 2025, 10am, University of Évry, Pelvoux Site, Yasmina Bestaoui Bx30 Amphitheather.
Title : Exploring Vision-Based Estimation and Riders’ Gaze Behavior to Improve Motorcycle Safety and Assistance.
Abstract
Motorcyclists remain disproportionately vulnerable in road traffic, motivating rider-centric safety advances that are affordable and deployable at scale. This dissertation tackles that need along two intertwined fronts: empirical studies that clarify how riders use visual information in curves, and monocular camera pipelines that estimate motorcycle dynamics in real time.
On the human-factors side, the thesis examines rider gaze across complementary contexts: an on-road comparison of gaze strategies in identical bends when driving a car versus riding a motorcycle; a virtual-reality paradigm that restricts visual access around the tangent point using circular apertures and targeted occlusions to quantify effects on steering and oculomotor behavior; and a VR evaluation of a gaze-guided visual cue designed to support novices in curves. Findings reinforce preview-based steering, illuminate the functional role of the tangent point, and reveal expertise-linked differences in scanning breadth.
The core of the technical contribution is a model-free and drift-free perception pipeline that operates without inertial sensors or complex vehicle models. This pipeline features an Inverse Perspective Mapping (IPM) method for estimating the full vehicle attitude relative to the road. To overcome the computational costs of IPM, a more efficient Direct Pixel Mapping (DPM) technique is introduced, which operates directly on image pixels to achieve superior performance. Furthermore, a geometric approach is developed to estimate lateral dynamics, including the side-slip angle and lateral velocity, by analyzing vanishing points and lane geometry.
The human-factors studies confirm that motorcycling imposes a higher cognitive load than car driving, compelling riders to adopt distinct gaze strategies focused on the tangent point and immediate path. The vision-based algorithms, validated through extensive simulation, demonstrate high accuracy (sub-degree for attitude) and real-time viability on commodity hardware. This research successfully demonstrates that a low-cost monocular camera can serve as a comprehensive sensor for motorcycle dynamics. The resulting algorithms and insights provide a critical foundation for creating accessible, intelligent ARAS that enhance situational awareness by aligning machine perception with rider gaze behavior.
Doctoral thesis jury composition
| Jury member | Title | Affiliation | Function in the jury |
|---|---|---|---|
| Hichem ARIOUI | Full Professor | Université Évry Paris-Saclay | Thesis Co-supervisor |
| Marie BABEL | Full Professor | INSA de Rennes | Examiner |
| Amine CHELLALI | Associate Professor | Université Évry Paris-Saclay | Thesis Supervisor |
| Michèle GOUIFFES | Full Professor | Université Paris-Saclay | Examiner |
| Régis LOBJOIS | Research Scientist with HDR | Université Gustave Eiffel | Reviewer |
| Rochdi MERZOUKI | Full Professor | Université de Lille | Reviewer |
| Pascal VASSEUR | Full Professor | Université de Picardie Jules Verne | Examiner |
- Date: Tuesday, December 9, 2025, 10 a.m.
- Location: University of Évry, Pelvoux Campus, Yasmina Bestaoui Lecture Hall Bx30, 36 rue du Pelvoux, 91080 ÉVRY-COURCOURONNES
- Doctoral student: Obaida ALRAZOUK (University of Évry Paris-Saclay, IBISC IRA2 team)
- Thesis supervision: Amine CHELLALI (Assistant Professor, University of Évry, IBISC IRA2 team), thesis supervisor; Hichem ARIOUI (Full Professor, University of Évry, IBISC IRA2 team), co-supervisor
