Management and control of crowds is crucial to maintaining public safety and is considered to be an increasingly relevant research topic in the social sciences. Developing a robust crowd control system CMS is a challenging task as it involves addressing several key issues such as density estimation, irregular flow of noise, occlu- sion, position estimation, etc.The development of a robust crowd control system CMS is a challenging task. This is because it involves ad- dressing several key issues such as density estimation, irregular flow of noise, occlusion, position estimation etc. Crowds that gather at places such as hospitals, cultural and religious centers are usually monitored with closed circuit television (CCTV) cameras to identify crowds and to mon- itor the crowd flow in order to reduce the damage. As a result, many researchers have turned to computer vision and machine learning to solve these issues by reducing the amount of human participation needed. This project presents a system for estimating the size of heterogeneous crowds made up of pedestrians walking in different directions without using segmentation or explicit object tracking. The system is based on the development of two algorithms, where the first algorithm YOLOV3 initially determines the objects with frames and classifies them, and then the movement of objects is predicted to determine the path through the second algorithm Deep SORT Crowd management and control is critical to maintaining public safety and is an important topic of research. Developing a robust crowd control system CMS is a challenging task as it involves addressing several key issues such as density estimation, irregular flow of noise, occlu- sion, position estimation, etc. Crowds gather in various places such as hospitals, or cultural and religious points are usually monitored by closed circuit television (CCTV) cameras to identify crowds and monitor crowd flow to reduce damage. Many researchers have turned towards com- puter vision and machine learning that overcomes these issues by reducing the need for human participation. We present a system for estimating the size of heterogeneous crowds, made up of pedestrians traveling in different directions, without using segmentation or explicit object tracking. This system depends on the development of two algorithms, where the first algorithm (YOLOV3) in turn determines objects with frames and classifies them, then the movement of objects is predicted to determine the path through the second algorithm Deep SORT.