Video surveillance is a vital part of smart cities, providing continuous monitoring and evaluation of metropolitan areas to improve security, safety, and efficiency. It uses strategically positioned cameras and associated technology to gather and analyze video footage, allowing for preemptive reactions to emergencies and other crucial occurrences. In this study, we have used deep learning algorithms for a comparative analysis of video surveillance in smart cities. Deep learning revolutionizes video surveillance by enabling intelligent systems to evaluate video data in real time, recognizing abnormalities, objects, and behaviors, resulting in more precise and efficient security procedures. The best architecture for object recognition in deep learning is CNN. This review will utilize various algorithms for comparison, as provided by CNN. In this study, the analysis of video surveillance systems is compared using well-known algorithms. This review will propose one of the most effective algorithms for video surveillance in smart cities following the comparison of the algorithms.