AI-Driven Quarrel Detection in Surveillance Footage

This paper presents a comprehensive study to improve violence detection capabilities in video footage, with a particular focus on the Violence Detection Dataset. The dataset comprises two classes, “safe” and “unsafe,” capturing a diverse range of scenarios. This research explores and compares the efficacy of two prominent machine learning algorithms, Random Forest and Convolutional Neural Network (CNN), in the real-time identification and categorization of safe and unsafe instances. After thorough evaluation, Random Forest is shown to be more accurate and efficient than CNN in detecting acts of violence. In addition to addressing issues with different visual datasets, the study extends beyond algorithmic comparison and provides insightful information about algorithm performance in a range of contexts. The effective use of Random Forest in realtime surveillance has great potential for public safety, law enforcement, and crowd management. The findings contribute not only to the evolving field of violence detection but also provide practical guidance for the implementation of machine learning algorithms in real-world surveillance scenarios. The outcomes of this study are significant for the creation of safer public areas by promoting the development of monitoring systems that are more efficient and secure. For scholars, professionals, and officials looking to enhance violence detection technologies, this report is a valuable resource.