Skip navigation
Please use this identifier to cite or link to this item: https://repository.esi-sba.dz/jspui/handle/123456789/816
Title: Acquisition and analysis of a dataset on the perception of walkability in the city
Authors: MENADI, ABderrahim
Keywords: walkability
Eye Tracking
Data Collection
Data Analytics
Fixations
Computer Vision
Semantic Segmentation
Gaze Analysis
Urban Perception.
Issue Date: 2025
Abstract: Urban walkability depends on both the physical characteristics of public spaces and how pedestrians visually perceive and explore their environment while walking. This thesis therefore combines subjective evaluation of pedestrian routes with objective analysis of eye movements to study, in situ, the relationship between visual perception and the quality of walking spaces. Nine participants walked a predefined route while wearing eye-tracking glasses. For each segment of the route, they rated the walkability and provided detailed comments. Simultaneously, a reproducible protocol synchronized fixations and saccades with each video frame, performed semantic segmentation of each frame, and assigned each gaze point to the corresponding visual object class (sidewalks, vehicles, buildings, vegetation, etc.). From this data, various indicators were calculated: proportion of fixations per object category, fixation and saccade durations, transition probabilities, gaze entropy, and visual diversity measures. These metrics were then aggregated per participant and per route segment to study their relationship with the collected subjective evaluations. This approach allows identifying the urban elements that attract the most attention and analyzing how gaze dynamics correlates with perceptions of comfort and overall walking experience. The results show which parts of the city consistently catch the attention of pedestrians and how eye movement patterns are linked to how walkable a place seems, both for different people and along different parts of the route. The thesis underscores the ramifications of these findings for pedestrian-oriented urban design and examines the study’s constraints, notably the limited sample size and the technical difficulties encountered in real-world eye-tracking applications. Lastly, the full dataset structure and the analytical pipeline are made public so that other researchers can use them and so that urban planners can make decisions based on how people feel about things**** Cette thèse examine la perception visuelle des rues à pied. Neuf volontaires ont parcouru un court trajet urbain à Nantes, équipés de lunettes d’oculométrie, et ont évalué chaque segment du trajet. Nous avons synchronisé les données d’oculométrie avec des images vidéo de la scène, appliqué une segmentation sémantique pour identifier les éléments visibles (trottoir, végétation, bâtiments, véhicules, etc.) et analysé où leur attention était portée. Les segments perçus comme plus praticables ont attiré davantage le regard vers la végétation et un trottoir dégagé, tandis que les segments moins bien notés ont focalisé l’attention sur les façades des bâtiments, la chaussée et la circulation. Les segments les mieux notés ont également montré des taux de saccades légèrement plus élevés, suggérant une exploration visuelle plus large. Ce projet propose une méthodologie et un ensemble de données robustes et reproductibles qui relient le comportement objectif du regard à l’expérience subjective de l’utilisateur. Ces outils peuvent être utilisés pour soutenir l’analyse urbaine basée sur la perception et fournir des conseils pratiques de conception pour créer des rues plus confortables et conviviales pour les piétons. iii
Description: Main Supervisor :Pr. Myriam SERVIÈRES /Main Supervisor : Pr. Sidi Mohamed BENSLIMANE /Co-Supervisor :Dr. Vincent TOURRE / Co-Supervisor : Dr. Jean-Philippe RIVIÈRE
URI: https://repository.esi-sba.dz/jspui/handle/123456789/816
Appears in Collections:Master

Files in This Item:
File Description SizeFormat 
Menadi-Master-1-1.pdf136,32 kBAdobe PDFView/Open
Show full item record


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.