Open Access

ARTICLE

Deep Trajectory Classification Model for Congestion Detection in Human Crowds

by Emad Felemban¹, Sultan Daud Khan², Atif Naseer³, Faizan Ur Rehman^4,*, Saleh Basalamah¹

1 Department of Computer Engineering, College of Computing and Information Systems, Umm Al-Qura University, Makkah, Saudi Arabia
2 Department of Computer Science, National University of Technology, Islamabad, Pakistan
3 Science and Technology Unit, Umm Al-Qura University, Makkah, Saudi Arabia
4 Institute of Consulting Research and Studies, Umm Al-Qura University, Makkah, Saudi Arabia

* Corresponding Author: Faizan Ur Rehman. Email:

Computers, Materials & Continua 2021, 68(1), 705-725. https://doi.org/10.32604/cmc.2021.015085

Received 06 November 2020; Accepted 02 February 2021; Issue published 22 March 2021

Abstract

In high-density gatherings, crowd disasters frequently occur despite all the safety measures. Timely detection of congestion in human crowds using automated analysis of video footage can prevent crowd disasters. Recent work on the prevention of crowd disasters has been based on manual analysis of video footage. Some methods also measure crowd congestion by estimating crowd density. However, crowd density alone cannot provide reliable information about congestion. This paper proposes a deep learning framework for automated crowd congestion detection that leverages pedestrian trajectories. The proposed framework divided the input video into several temporal segments. We then extracted dense trajectories from each temporal segment and converted these into a spatio-temporal image without losing information. A classification model based on convolutional neural networks was then trained using spatio-temporal images. Next, we generated a score map by encoding each point trajectory with its respective class score. After this, we obtained the congested regions by employing the non-maximum suppression method on the score map. Finally, we demonstrated the proposed framework’s effectiveness by performing a series of experiments on challenging video sequences.

---

Keywords

---