<![CDATA[Pedestrian areas often serve as centers of high public activity, requiring intelligent monitoring systems to ensure the safety and comfort of their users. The application of computer vision technology, particularly object detection, offers a promising approach for identifying and estimating the number of individuals in open public spaces. This study implements the YOLOv8 algorithm to develop a human detection and crowd counting model within the pedestrian zones of the Bogor Botanical Garden. Data were collected in the form of images and videos from three strategic locations and annotated using Roboflow with a single object class labeled "person.” The model was trained on the Google Colab platform using a Region of Interest (ROI)-based approach and evaluated through confusion matrix, precision, recall, F1-score, and mean Average Precision (mAP). Results indicate a precision of 0.846, recall of 0.858, F1-score of 0.85, and mAP@50 of 0.951, although a performance drop was observed at mAP@50-95 with a score of 0.586. These findings suggest that YOLOv8 demonstrates strong real-time performance in pedestrian human detection, while challenges remain in enhancing precision under complex and varied conditions.]]>
Copyrights © 2025
