Autonomous crowd activity analysis is imperative in light of recent global events. The study of crowd behavior has generated considerable interest in the fields of computer vision and cognitive science. Consider a scenario where a security guard is tasked with monitoring a multitude of CCTV camera feeds, particularly those used in police investigations. To effectively manage, count, secure, and track a crowd occupying a shared space, such as during COVID-19 outbreaks and public events. Assessing crowd situations presents challenges due to substantial occlusion, intricate actions, and variations in posture. The development of an algorithm capable of monitoring each video and automatically identifying suspicious activity would enhance the accuracy and efficiency of a large-scale intelligent system. This serves as the driving force behind video anomaly detection. The fundamental concept involves training a model on typical activities using training videos featuring normal behavior, and subsequently utilizing those activities that are different from any normalized seen in the training video. The comprehensive inclusion of all conceivable future anomalous events within the training video is unattainable due to the inherent limitations associated with the complete knowledge and capture of such occurrences. Deep learning, a branch of machine learning, employs artificial neural networks to acquire knowledge from data. Notably, deep learning has been successfully employed in the field of crowd anomaly detection, which poses significant challenges but holds great significance. Deep learning has demonstrated considerable potential for addressing this problem and has surpassed conventional machine learning systems in terms of performance. This review aims to present an exhaustive analysis of video anomaly detection systems based on deep learning techniques that were introduced since year 2019.