Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Stampede Alert Clustering Algorithmic System Based on Tiny-Scale Strengthened DETR

Mingze Sun, Yiqing Wang, Zhenyi Zhao

TL;DR

The paper tackles stamping-and-crowd safety in dense environments by extending Deformable DETR with a tiny-scale strengthened framework that enhances small-target detection while preserving speed. It introduces a multi-scale deformable attention mechanism and a dedicated tiny-scale feature fusion module, supplemented by SwiGLU in the FFN to accelerate training and improve generalization. A semi-supervised stamping-prediction pipeline based on K-means and KNN clustering analyzes crowd density and congestion to provide early warnings. Experimental results on the PKSLHR and related datasets indicate improved small-object detection performance and practical viability for real-time safety monitoring in airports and train stations. The work demonstrates meaningful gains in precision for micro-targets and offers a robust framework for crowd safety analytics that combines vision-based detection with congestion-aware risk forecasting.

Abstract

A novel crowd stampede detection and prediction algorithm based on Deformable DETR is proposed to address the challenges of detecting a large number of small targets and target occlusion in crowded airport and train station environments. In terms of model design, the algorithm incorporates a multi-scale feature fusion module to enlarge the receptive field and enhance the detection capability of small targets. Furthermore, the deformable attention mechanism is improved to reduce missed detections and false alarms for critical targets. Additionally, a new algorithm is innovatively introduced for stampede event prediction and visualization. Experimental evaluations on the PKX-LHR dataset demonstrate that the enhanced algorithm achieves a 34% performance in small target detection accuracy while maintaining the original detection speed.

Stampede Alert Clustering Algorithmic System Based on Tiny-Scale Strengthened DETR

TL;DR

The paper tackles stamping-and-crowd safety in dense environments by extending Deformable DETR with a tiny-scale strengthened framework that enhances small-target detection while preserving speed. It introduces a multi-scale deformable attention mechanism and a dedicated tiny-scale feature fusion module, supplemented by SwiGLU in the FFN to accelerate training and improve generalization. A semi-supervised stamping-prediction pipeline based on K-means and KNN clustering analyzes crowd density and congestion to provide early warnings. Experimental results on the PKSLHR and related datasets indicate improved small-object detection performance and practical viability for real-time safety monitoring in airports and train stations. The work demonstrates meaningful gains in precision for micro-targets and offers a robust framework for crowd safety analytics that combines vision-based detection with congestion-aware risk forecasting.

Abstract

A novel crowd stampede detection and prediction algorithm based on Deformable DETR is proposed to address the challenges of detecting a large number of small targets and target occlusion in crowded airport and train station environments. In terms of model design, the algorithm incorporates a multi-scale feature fusion module to enlarge the receptive field and enhance the detection capability of small targets. Furthermore, the deformable attention mechanism is improved to reduce missed detections and false alarms for critical targets. Additionally, a new algorithm is innovatively introduced for stampede event prediction and visualization. Experimental evaluations on the PKX-LHR dataset demonstrate that the enhanced algorithm achieves a 34% performance in small target detection accuracy while maintaining the original detection speed.
Paper Structure (20 sections, 25 equations, 7 figures, 1 table)

This paper contains 20 sections, 25 equations, 7 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Works
  3. Feature Fusion
  4. Multi-scale Deformable Multi-head Attention Module
  5. Transformer-based DETR
  6. Deformable DETR Target Detection Model
  7. The Design, Intent, and Limitations of the Templates
  8. The Core Architecture of the DETR Object Detection Model
  9. Introducing the SwiGLU Activation Function in FFN
  10. Self-Attention
  11. Multi-Head Self-Attention
  12. The Training Process
  13. Experimental environment and data set
  14. Evaluation Criteria - IoU
  15. Experimental Results
  16. ...and 5 more sections

Figures (7)

  • Figure 1: Architecture of Tiny-Scale Strengthened DETR
  • Figure 2: Comparison of several loss functions
  • Figure 3: Decision boundary of the activation function
  • Figure 4: Multi-head attention mechanism based on PKSLHR data set
  • Figure 5: Algorithm overall frame diagram
  • ...and 2 more figures