CCTV camera systems are widely used to monitor public spaces but generally
only record without producing data that can be analyzed further. In fact, accurate
data is really needed to support evaluation and improvement of public services.
This research aims to apply the YOLO (You Only Look Once) algorithm to build a
system capable of counting the number and duration of visitors. This research was
conducted at a train station using CCTV footage as input. The system detects human
objects, assigns a unique ID, and records attendance time into a CSV file for further
analysis. The initial training process was conducted using a private dataset, but in
the final implementation, the pre-trained YOLO model was used because it provided
more stable performance in real-world environments. Evaluation shows that the
system works well in low-movement environments but experiences a decrease in
accuracy when objects move quickly or overlap. With a minimum duration
threshold of 3 seconds, the system is able to improve tracking consistency. This
research proves that the computer vision approach can generate quantitative data
useful in supporting data-driven decision-making in public spaces.