Integrasi Algoritma YOLOv8 dan Streamlit untuk Visualisasi Real-Time dan Akurat dalam Penghitungan Kerumunan di Kawasan Stasiun Bekasi

Abstract

Crowd management in public transportation areas has become a critical challenge with the rise of urban populations. This study develops a real-time web-based people detection and counting system by integrating the YOLOv8 algorithm with the Streamlit framework. A case study was conducted at the entrance of Bekasi Station. The model was developed using the AI Project Life Cycle approach, and the system was built following the Waterfall methodology. Data were obtained from video recordings, which were extracted into images, annotated, and processed into training and testing datasets. The YOLOv8 model was trained for 50 epochs, yielding strong performance with an mAP@0.5 of 91.7%, a maximum precision of 93.6%, and an F1-score of 87%. Tests on 15 images showed an average accuracy of 80.37% and an error rate of 19.63%. The model's performance declined on out-of-dataset images due to variations in lighting and extreme crowd density. The system was tested using black-box testing and demonstrated that all main features—image upload, object detection, visualization, and result download—functioned correctly. The system has been successfully deployed on Streamlit Cloud. These results indicate that the system offers a practical, lightweight, and responsive solution to support crowd monitoring in public areas. In future development phases, the system can be extended to support real-time video stream processing and integrated with an object tracking and classification module to accurately identify and differentiate the ingress and egress flow of individuals within a defined surveillance area.