%0 Conference Paper
%A Alia, Ahmed
%A Chraibi, Mohcine
%A Seyfried, Armin
%T A Dynamic Distance Social LSTM for Predicting Pedestrian Trajectories in Crowded Environments
%M FZJ-2025-03048
%D 2025
%X This work introduces dynamic distance Social Long Short-Term Memory, a deep learning approach for pedestrian trajectory prediction in crowded environments. The approach integrates a new dynamic distance-based loss function into Social Long Short-Term Memory, enhancing collision avoidance without compromising displacement accuracy. The method is trained and evaluated on a heterogeneous density dataset and four homogeneous density datasets, covering various crowd-density levels. Experimental results show that the proposed approach outperforms baseline methods in reducing collision rates without decreasing displacement accuracy and, in most cases, even improving it.
%B Modelling, Data Analytics and AI in Engineering
%C 7 Jul 2025 - 11 Jul 2025, Porto (Portugal)
Y2 7 Jul 2025 - 11 Jul 2025
M2 Porto, Portugal
%F PUB:(DE-HGF)6
%9 Conference Presentation
%R 10.34734/FZJ-2025-03048
%U https://juser.fz-juelich.de/record/1044146