Simulasi Dinamis Okupansi Parkir Menggunakan Deteksi Ruang Parkir Berbasis Computer Vision

Abstract

Parking facilities are difficult to manage when the number of arriving vehicles grows, and knowing the current parking condition does not explain how occupancy will change under higher demand. This study developed a framework that integrates YOLO26s-based parking space detection with a stochastic occupancy simulation. On the Roboflow test set, the detector achieved a precision of 93.6%, recall of 91.8%, and mAP@50 of 95.5%, although remaining false positives and false negatives still affect the simulation input. Applied to the parking video, the median detections of the first ten frames gave an estimated capacity of 174 spaces and an initial occupancy of 58.6%. These values are estimates, as the detected capacity varied between 150 and 194 spaces across frames. The simulation of three arrival scenarios showed that the low scenario stayed below capacity, with a mean occupancy of 50.78%, whereas the moderate and high scenarios reached full capacity in all runs after mean times of 124.5 and 30.5 minutes. A higher arrival rate therefore brings saturation forward and lengthens the time spent at capacity. However, the arrival and departure parameters are scenario assumptions, so the results are not forecasts of real parking demand. Future work should use longer videos, validate capacity against ground truth, and add vehicle tracking to estimate arrival and departure rates.