Peningkatan YOLO11 dalam Melakukan People Tracking and Counting Based On Gender

Abstract

This research is motivated by the need for an intelligent video-based monitoring system capable of detecting, tracking, and counting people based on gender accurately and in real-time. Manual methods for counting visitors are often inefficient and prone to errors, especially in crowded environments with high object movement dynamics. To address this issue, a Computer Vision approach based on Deep Learning was implemented by modifying the YOLO11 architecture using the C2f (Cross Stage Partial with Fusion) module and integrating it with the ByteTrack algorithm for object tracking. The main objective of this research is to analyze and compare the performance of the original YOLO11 model and the modified YOLO11 C2f in detecting and tracking human movement by gender in video data. The research methodology includes model training with variations of optimizers (AdamW and SGD) and different batch sizes, followed by evaluation using metrics such as Precision, Recall, F1-Score, mAP50, and mAP50–95, as well as analysis of object tracking performance. The experimental results show that the YOLO11m C2f model using the SGD optimizer and a batch size of 32 achieved the best performance with a Precision of 0.983, Recall of 0.976, F1-Score of 0.9799, mAP50 of 0.992, mAP50–95 of 0.926, and accuracy of 95.7%. Although its accuracy is slightly lower than the YOLO11m AdamW (95.73%), the C2f model demonstrated higher stability in real-time tracking and counting, effectively reducing identity switching between male and female classes. In conclusion, the integration of the YOLO11 C2f architecture with the ByteTrack algorithm provides an efficient, accurate, and stable detection and tracking system suitable for real-time people tracking and counting applications in dynamic environments.