Advancements in Suspicious and Violent Activity Recognition for Intelligent Video Surveillance

Recognition of suspicious or violent activity in video surveillance has become increasingly important in terms of public safety and security. This synthesis examined state-of-the-art methodologies for detecting anomalous human behaviors-such as machine learning, deep learning, and hybrid approaches. By analyzing key contributions across recent studies, we identify advancements in feature extraction, model architectures such as CNNs, LSTMs, and Conv3D, and their applications to datasets like UCF101 and custom video repositories. The comparative analysis shows promising aspects in terms of performance and reliability, with some techniques such as Time-Distributed CNN getting accuracy improvements of over 90%. The issues of practical applications in a real-world scenario, problems with dealing with advanced datasets, and future research directions such as real-time implementation and ethical considerations were discussed. The present study provides an inclusive overview of the current trends in intelligent surveillance systems and the roadmap for innovations in the near future.