DeepCAMS: A Deep Learning Approach for Real-Time Crowd Monitoring and Suspicious Behavior Detection Using Spatial-Temporal Analysis
Ayman A. Alharbi · Engineering Technology & Applied Science Research · 2025
The increasing need for robust and intelligent crowd monitoring systems has led to advances in deep learning-based solutions. However, existing methods often struggle with capturing complex crowd dynamics and detecting suspicious behaviors in real-time. This study introduces DeepCAMS (Deep Learning-based Crowd Analysis and Monitoring System), a novel architecture that integrates a Fully Convolutional Network (FCN) for spatial feature extraction and a Long Short-Term Memory (LSTM) network for temporal analysis. Unlike traditional methods, DeepCAMS addresses the limitations of static and shallow models by combining spatial and temporal insights, enabling accurate classification of crowd behaviors as Normal or Suspicious. DeepCAMS demonstrated superior performance across multiple metrics, marking a substantial improvement over traditional approaches. The ability of DeepCAMS to adapt to diverse crowd densities and identify subtle behavioral anomalies highlights its scalability and practical application in real-world surveillance. Therefore, DeepCAMS sets a new benchmark in crowd behavior analysis by offering a unified spatial-temporal framework that ensures high accuracy, adaptability, and efficiency in dynamic environments. This study not only advances the field of smart surveillance but also paves the way for future research on scalable and interpretable crowd monitoring systems.