<![CDATA[This study presents the implementation of the YOLOv8 algorithm to enhance real-time crowd counting on the ngedatedotid application, which aims to provide accurate crowd density information at various locations. The proposed model leverages the advanced capabilities of YOLOv8 in detecting and localizing head-people objects within crowded environments, even in complex visual conditions. The model achieved a mAP of 85%, outperforming previous models such as YOLO V8'S (78.3%) and YOLO V7 (81.9%), demonstrating significant improvements in detection accuracy and localization capabilities. The custom-trained model further exhibited a detection accuracy of up to 95% in specific scenarios, ensuring reliable and real-time feedback to users regarding crowd conditions at various locations. By implementing a microservices architecture integrated with RESTful API communication, the system facilitates efficient data processing and supports a modular approach in system development, enabling seamless updates and scalability. This architecture allows for independent deployment of services, thereby minimizing system downtime and optimizing performance. The integration of YOLOv8 and the custom-trained model has proven to be effective in enhancing real-time monitoring and detection of crowd density, making it a suitable solution for diverse applications that require dynamic and accurate crowd information. The results indicate that the proposed model and system architecture can provide a robust framework for real-time crowd management, which is crucial for business owners, event organizers, and public safety monitoring. Future research should consider exploring newer versions of YOLO, such as YOLO V9-S, and expanding the dataset to address challenges related to varying lighting conditions, occlusions, and object orientations. Optimizing these factors will further improve the model’s accuracy and reliability, setting a new standard for crowd detection systems in public spaces and enhancing the overall user experience.]]>
