DECODING THE CROWD: HIGH-ACCURACY INDIVIDUAL EMOTION IDENTIFICATION THROUGH SIMULATED BEHAVIOR ANALYSIS

Syed Zohair Quain Haider; Abdulrehman Arif; Muhammad Zeeshan Haider Ali; Muhammad Ans khalid

Authors

Syed Zohair Quain Haider Department of Computer science and Information Technology, University of Southern Punjab, Multan, Pakistan
Abdulrehman Arif Department of Computer Science and Information Technology, University of Southern Punjab, Multan.
Muhammad Zeeshan Haider Ali Department of Computer Science and Information Technology, University of Southern Punjab, Multan.
Muhammad Ans khalid Department of Computer Science and Information Technology, University of Southern Punjab, Multan.

Keywords:

Facial Expression Recognition, Support Vector Machine (SVM), Quadratic SVM Classifier, Emotion Detection, Crowd Emotion Analysis, Feature Extraction, HOG Features (Histogram of Oriented Gradients), Machine Learning Model Training, Image-based Emotion Detection

Abstract

Research into understanding and quantifying crowd emotions has gained significant importance in recent years, driven by applications ranging from public safety and event management to urban planning and social psychology. This study proposes a novel computational framework specifically designed to identify and analyze the individual emotional states that collectively shape crowd dynamics. We define emotions as dynamic, evolving responses triggered by complex interactions between internal cognitive states and external stimuli within unfolding events. While traditional observational methods rely primarily on camera-based facial recognition to detect aggregate emotional expressions at the crowd level, our framework advances the field by focusing on granular, individual-level emotion identification within the collective. This is achieved through a sophisticated simulation engine that models diverse crowd scenarios and behaviors. The model explicitly evaluates the bidirectional influence between individual emotional states and emergent crowd behavior, allowing us to categorize distinct types of crowd dynamics (e.g., panic, cohesion, dispersion) based on underlying emotional profiles derived from the analysis. Our approach integrates simulated facial expression data (representing individuals) with contextual event triggers and simulated crowd movement patterns. Key contributions include: (1) A validated method for inferring individual emotions within dense crowds using advanced pattern recognition adapted to simulated sensor data; (2) Quantitative analysis demonstrating how specific emotional valences (e.g., fear, excitement) propagate and influence collective behavior trajectories; and (3) A classification system for crowd dynamics grounded in real-time emotion analysis. Rigorous validation using synthetic datasets and benchmarks shows the proposed system achieves a statistically significant improvement (15-20% higher F1-score) in emotion detection accuracy compared to existing state-of-the-art crowd-level recognition methods. This enhanced capability holds substantial promise for developing more responsive crowd management systems and predictive models for large gatherings.