<![CDATA[The provision of accurate and easily accessible public information regarding the condition of urban parks remains a challenge in the management of public open spaces. People still do not have a system that allows them to monitor park crowd levels directly, making real-time observation of park conditions difficult. Previous studies on crowd counting have mainly focused on improving object detection performance, while the implementation of crowd monitoring systems as public information services remains limited. Therefore, this research integrates the YOLOv11 algorithm with a microservices architecture to provide real-time crowd information through a public park monitoring website. This research develops a park visitor counting information system based on computer vision using the YOLOv11 algorithm to detect and count crowds from CCTV video streams. The system is designed using a microservices architecture with the FastAPI framework to support real-time detection and data integration into the Surabaya park monitoring website. The research process involves several stages, including dataset preparation, data labeling using Roboflow, YOLOv11 model training, selection of the most optimal optimizer, and implementation of the system on the detection backend. The results show that the YOLOv11m model with the SGD optimizer achieved the best performance, obtaining an mAP@50 score of 92.76%, a recall value of 89.75%, and an F1-score of 90.07%. In addition, the system successfully performed real-time crowd detection and counting under various crowd density levels, lighting conditions, and CCTV camera angles.]]>
