Detecting Abusive Activity with Enhanced Convolutional Neural Networks
Suadad Zaidan Khalaf, Mohamed Ibrahim Shujaa, Ahmed Bahaaulddin A. Alwahhab · 2023
There is a growing reliance on surveillance systems for law enforcement and public safety purposes. Video surveillance scenarios, such as those seen in train stations, schools, and hospitals, must automatically detect hostile and suspicious behavior to avoid social, economic, and environmental harm. Human contact can now be seen using intelligent video surveillance systems. Crowds and the camera’s field of view reduce the system’s effectiveness as a security procedure. As a result, there is a lot of study on the many strategies for detecting violent behavior. The method of detection discussed in this study can be divided into three categories. These groups were established as a result of the classification procedures that were used. Among the categorizations is a refined CNN for using AI to identify instances of violence. Principal Component Analysis (PCA) is used to develop techniques for feature extraction and violence detection. The AIRTLab dataset was utilized to analyze models based on dataset learning for violence detection by testing the resilience of algorithms against false positives. To evaluate the proposed model’s robustness against false positives, we calculated accuracy measures and utilized them to establish a baseline on the AIRTLab dataset. The study literature claims a 99.98% success rate for the suggested models, with transfer learning-based networks boasting greater generalization abilities.