Real-World Security Applications Through Computer Vision
Asish Kumar Dalai, Hitesh Mohapatra · 2025
The model is trained on a diverse dataset that includes various scenarios of violent interactions, allowing it to perform consistently across different surveillance environments and adapt to changing conditions. The chapter details the training process, including the data augmentation techniques employed to improve the model's generalization. Additionally, the system incorporates behavioral analysis to minimize false positives, helping to differentiate between brief aggressive actions and normal behavior. Extensive testing using real-world surveillance footage demonstrates the system's accuracy in detecting and classifying violent events. Comparative analysis shows that the solution outperforms existing methods in terms of both precision and recall. The chapter also explores the practical implications of deploying such a system, including its scalability for large monitoring networks and its potential integration with existing security infrastructures.