COGNITIVE SURVEILLANCE USING DEEP NEURAL NETWORKS FOR PREDICTING ROBBERY TENDENCIES IN INDOOR SCENARIOS

Abstract

Ensuring safety in indoor environments such as banks, offices, and retail spaces has become increasingly critical with the rise in security threats and suspicious human activities. Traditional surveillance systems are primarily reactive, relying on manual monitoring and post-event analysis, which limits their ability to prevent incidents in real time. This research proposes a cognitive surveillance framework that leverages deep neural networks to predict robbery tendencies based on human behavioral cues captured through indoor security cameras. The proposed system integrates computer vision, motion analysis, and behavioral modeling to extract spatial-temporal features indicative of potential criminal intent. Using convolutional and recurrent neural network architectures, the framework learns subtle pre-incident behavioral patterns such as abnormal movement trajectories, prolonged loitering, or concealed object handling. A behavioral intent recognition module is employed to evaluate the probability of a robbery attempt before it occurs, thereby enabling proactive threat intervention. Extensive experiments conducted on custom-labeled indoor surveillance datasets demonstrate that the proposed model achieves high accuracy in distinguishing normal and suspicious activities. The system also exhibits robust generalization across varying lighting conditions, crowd densities, and camera angles. By transforming traditional surveillance into a predictive and intelligent security mechanism, this approach enhances situational awareness and empowers automated decision-making in real-time security operations.