Violations of helmet use regulations by motorcyclists in the Universitas Riau campus area still frequently occur. Supported by a preliminary survey indicating that 98.6% of 149 respondents perceive campus supervision as highly lenient, this lack of direct supervision creates a gap where students tend to ignore safety rules. This study aims to develop an automated helmet detection system using the YOLOv8 deep learning algorithm to classify riders. The research process was structured following the CRISP-DM (Cross-Industry Standard Process for Data Mining) framework. During the modeling phase, the algorithm was trained utilizing a combined dataset of 4,321 augmented images. The dataset comprises primary data from campus observations and secondary data from the Roboflow public dataset, functioning as a domain adaptation strategy. Evaluation of five architectural variants showed that YOLOv8n (Nano) provided the most optimal performance, successfully achieving a mean Average Precision (mAP@0.5) of 91.43% with a highly compact weight size of 11.9 MB. This model was then implemented into a web-based interface using the Flask framework for real-time monitoring. Field testing conducted at three campus entrance gates against 179 individuals demonstrated a robust operational accuracy of 89.4%, processing video streams at an average of 5 to 5.5 Frames Per Second (FPS). Nevertheless, occasional object tracking failures (lost tracking) were encountered due to motion blur effects when vehicles traveled at high speeds, primarily caused by the limited resolution and FPS specifications of the built-in webcam used during the test.