Real-Time Violence Recognition in CCTV Video Surveillance Enhanced by AI Deep Technique
Narenthirakumar Appavu · 2025
Despite its many uses, deep learning (DL) is a key tool in the machine learning (ML) community's toolbox. Numerous techniques have been developed to identify violence and criminal activity on video since many crimes take place in public areas that are not adequately monitored. The significance of automatic violence detection in surveillance video studies has grown. They do, however, have several drawbacks and frequently rely on particular circumstances. This article presents a method for detecting aggression in surveillance films that is based on convolutional neural networks (CNNs). Accuracy is increased by applying both deep learning and machine learning methods as advised. Performance assessments showed how effectively the suggested approach identifies violent content in video. Test results show that the future solution works better than current approaches for classifying violent then criminal material in films. The models that were previously trained InceptionV3, ResNetV2, Inception_ResNetV2, Violence_Net, and the suggested vioNet remained trained on four datasets. The following were the classification outcomes of the endorsement data for each model: VioNet model 99.72%, InceptionResNetV2 90%, ViolenceNet pseudo-97, and InceptionV3 92%. Due to its easier integration of feature maps, the Dense_Net prototypical was selected for our application arrangement instead of the more intricate Inception or InceptionResNet models. Compared to earlier iterations, its sturdy construction provides greater accuracy and performance with fewer filters and settings.