Leveraging a Hybridized CNN-HARS Surveillance Model for Security Threat Detection in Imo State, Nigeria

The escalating security threats in Imo State, including terrorism, unknown gunmen attacks, kidnapping, one-chance drivers, and armed robbery, have created an urgent need for effective surveillance solutions. This study presents an integrated Convolutional Neural Network (CNN) and Human Activity Recognition System (HARS) model to enhance real-time security surveillance. Video feeds from strategically deployed CCTV cameras and drones are processed through a CNN for spatial feature extraction, identifying critical elements like edges, textures, and human shapes. Features indicative of suspicious behaviours, such as loitering or carrying weapons, are refined using Rectified Linear Unit (ReLU) activation functions and downsampled via pooling. Flattening and fully connected layers classify activities into normal (0.0–0.3), suspicious (0.4–0.6), or threat (0.7–1.0) categories. The HARS component complements CNN by performing temporal analysis, recognising activity patterns that may escalate into criminal behaviour. The system triggers real-time alerts for suspicious and threat activities, enabling rapid responses from security agencies. The model was evaluated on a 500-sample label dataset, achieving an accuracy of 84%, a precision of 86.96%, a recall of 80%, and an F1-score of 83.3%, demonstrating its capability to detect and classify security threats effectively. The findings highlight the transformative potential of advanced surveillance systems in improving public safety, aiding law enforcement, and restoring stability in Imo State. This innovative approach provides a robust tool for combating crime and enhancing security in both urban and remote areas.