Violent Activity Detection Through Surveillance Camera Using Deep Learning

Surveillance systems have been widely deployed across the globe to address the escalating prevalence of criminal activities. With the aim of enhancing public safety, the integration of computer vision has yielded increased sophistication, reliability, and efficiency within these systems. However, opportunities for improvement persist. Although contemporary systems proficiently capture incidents, their capacity to do so intelligently, facilitating prompt law enforcement responses to assist victims and prevent further criminal endeavors, remains nascent. In this paper, we present an in-depth exploration of methods to discern various forms of criminal conduct, including physical altercations, harassment, hijacking, and snatching, among others. Our approach leverages image processing principles employing distinct neural networks such as MobileNet-V2, ResNet50 V2, and ConvLSTM. These architectures are employed to correlate images with the pre-established trained system. The proposed model is tailored to identify specific criminal activities, including but not limited to punching, kicking, slapping, and instances of weapon-related violence. All pertinent data will be meticulously stored within the database for reference and analysis.