Recent global events have highlighted the need for autonomous crowd analysis. This technology is gaining interest in computer vision and cognitive science, as well as in the real world, where crowded scenes and their behavior are being increasingly studied. With the rise of crowd phenomena in the real world, crowded scene and their behaviour analysis has gotten a lot of attention recently. When overcrowded towns are subjected to regular crowded events such as strikes, protests, parades, any religious gathering or other types of public gatherings, they face a slew of security challenges. To address these concerns, human security personnel are often deployed to supervise meetings and protect the safety of attendees. However, in crowded spaces such as COVID-19 breakout sessions and public events, it can be challenging to manually manage, count, secure, and track everyone present. An automated system could help to improve safety and efficiency in these situations. Due to significant occlusion, complicated actions, and posture changes, assessing crowd situations is difficult. Crowd analysis is divided into two main categories: crowd statistics and crowd behavior analysis. Crowd statistics determines the level of service (LoS) of a crowded environment, while crowd behavior analysis identifies the mobility patterns and activities of people in a scene. Crowd behaviour analysis has become an essential tool for assuring peaceful event organizing and minimal casualties in public and religious venues across the world. Traditionally, handmade characteristics were used to compute crowd analysis. This paper provides an overview of contemporary exploration study of various crowd behaviour analysis approaches and address new unexplored problem with the eye of deep learning.