Video surveillance systems are crucial for mitigating crime and violence in public areas and events by observing individuals’ behavior and identifying normal or suspicious activities. Artificial intelligence has significantly advanced video processing by utilizing algorithms for detecting suspicious activities. This review paper examines and analyzes methodologies for identifying suspicious activities utilizing video surveillance systems. It emphasizes various frameworks for identifying suspicious activities through deep learning, machine learning, and conventional methods as the primary methodology. Furthermore, it encompasses an analysis of methodologies for assessing accuracy and identifying anomalies in real-time processing. This study also references video datasets that identify suspicious human activities in both crowded and uncrowded environments. Consequently, the study informs researchers of the most recent findings regarding the protection and security of public spaces, emphasizing key strengths and weaknesses in these technologies, thereby facilitating the development of new solutions to enhance the safety and security of citizens in public and densely populated areas.