Cote: |
1297 |
Auteur: |
HENKING Max |
Année: |
Janvier 2024 |
Titre: |
An advanced machine learning pipeline for pedestrian counting from video: performance comparison between different algorithms, model parameters and input sources |
Sous la direction de: |
Dr Christian Kaiser |
Type: |
Mémoire de master en géographie |
Pages: |
49 |
Complément: |
13 pages d'annexes paginées |
Fichier PDF: |
Mémoire [9.9 Mo]
|
Mots-clés: |
Pedestrian counting / pedestrian detection / pedestrian tracking / machine learning / spatial planning |
Résumé: |
Tracing and counting pedestrians is a growing area of interest in spatial analysis in the current context of fast growing and denser urban spaces. This research is part of this trend and presents an automated method that facilitates the extraction and analysis of pedestrian data from cameras for decision-support purposes in the field of spatial planning. There are two main focuses to this research: the first focuses on the extraction methods, which are tested by various combinations of algorithms and their associated parameters (detection threshold and tracking memory). The second assesses the reliability of these results by comparing algorithms and analysing raw images to validate these observations. The resulting optimised pipeline via the identification of algorithms, parameters and external factors improving pedestrian data extraction. |