Pedestrian dynamics modeling and social force analysis based on object detection

IntroductionObject detection is a fundamental component of modern computational applications, playing a crucial role in pedestrian analysis, autonomous navigation, and crowd monitoring. Despite its widespread utility, pedestrian-oriented object detection faces significant challenges, including dynam...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Frontiers in physics Ročník 13
Hlavní autoři: Yang, Fangyan, Liu, Rong, Zhu, Daoyu
Médium: Journal Article
Jazyk:angličtina
Vydáno: Frontiers Media S.A 07.07.2025
Témata:
crowd dynamics
deep learning
pedestrian detection
social force model
variational inference
ISSN:2296-424X, 2296-424X
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:IntroductionObject detection is a fundamental component of modern computational applications, playing a crucial role in pedestrian analysis, autonomous navigation, and crowd monitoring. Despite its widespread utility, pedestrian-oriented object detection faces significant challenges, including dynamic crowd behaviors, occlusions, multi-scale variability, and complex urban environments, which hinder the accuracy and robustness of existing models.MethodsTo address these challenges, we propose a novel framework that integrates the Information-Geometric Variational Inference Framework (IGVIF) with the Adaptive Exploration-Exploitation Trade-off Strategy (AEETS), specifically tailored for pedestrian dynamics. IGVIF formulates pedestrian detection as a probabilistic inference problem, leveraging principles from information geometry to efficiently explore high-dimensional parameter spaces. By incorporating techniques such as Riemannian optimization and multi-scale parameterization, IGVIF effectively captures the hierarchical and multi-modal structures inherent in pedestrian movement patterns. AEETS dynamically balances global exploration with local refinement using entropy-based metrics and feedback-driven adjustments, allowing the system to adaptively optimize complex loss landscapes with greater precision in pedestrian scenarios.ResultsTogether, these components create a robust and adaptive framework that overcomes traditional limitations by efficiently handling large-scale pedestrian variability and densely populated environments. Experimental evaluations across multiple real-world pedestrian datasets demonstrate the superiority of our physics-inspired approach, achieving state-of-the-art performance in pedestrian detection and movement analysis.DiscussionThis work highlights the transformative potential of interdisciplinary strategies in advancing pedestrian-aware object detection, bridging computational physics with deep learning methodologies to enhance urban mobility and crowd safety.
ISSN:2296-424X
2296-424X
DOI:10.3389/fphy.2025.1579280