This paper deals with the design and construction of a deep-learning-based human detection and counting system that is specifically designed for crowded places. The main goal is to monitor human presence within controlled regions, providing real-time supervision for applications including crowd control, intrusion detection, and retail analytics. To exploit the efficiency and accuracy of a YOLOv5 based model for object detection, this framework interprets image data in order to identify human shapes in highly cluttered, dynamic settings. The project faced various challenges, including extensive background noise, small object sizes, and limitations in handling real-time video input, which necessitated a shift to static image processing for performance optimization. This paper elaborates on the design, dataset preparation, model training, and testing processes, alongside an extensive review of related work. However, through overcoming the limitations of model training, label definition, and system adaptation, this work builds towards a scalable, robust human detection model that is applicable to a wide range of real-world scenarios, future prospects include multi-camera incorporation, activity identification, and edge-of-the-network applications.