<![CDATA[This study developed a real-time, computer vision-based traffic monitoring system to address safety concerns regarding uncontrolled public vehicle access within the Politeknik Negeri Bengkalis campus environment. As the existing CCTV infrastructure lacked intelligent analytical capabilities, an automated solution was required to extract objective traffic data. The system utilized the YOLOv8n model for vehicle detection and the ByteTrack algorithm for multi-object tracking to prevent duplicate counting across frames. To overcome classification inaccuracies inherent in pre-trained models without relying on resource-intensive retraining, a rule-based post-processing method was implemented. This method evaluated bounding box geometries, including pixel area, aspect ratio, and a vertical camera perspective factor, to filter raw detections. Evaluation results demonstrated that the rule-based approach significantly minimized false positives, achieving a total precision of 93.48% and an overall accuracy of 86.00%, which vastly outperformed both baseline and fine-tuned configurations. Furthermore, comparative tests across model variants confirmed that YOLOv8n provided the most stable CPU execution, maintaining the highest frame rate and lowest inference latency. The integrated system successfully tracked vehicle counts and estimated relative speeds, providing a reliable, GPU-independent analytical tool to support data-driven campus traffic management policies.]]>
