Deteksi Dan Perhitungan Kendaraan pada Pintu Akses Pariwisata Menggunakan Algoritma Yolov8 Deepsort

Abstract

Vehicle detection and counting at tourism access gates is a crucial task for managing visitor traffic and tourism infrastructure. Current methods are still conducted semi-manually, which is inefficient and prone to human error. This study aims to develop an automated vehicle detection and counting system by integrating the YOLOv8 algorithm with the DeepSORT tracking algorithm, enhanced with the Space-to-Depth Convolution (SPD-Conv) module. The dataset consists of 1,500 images of three vehicle classes — motorcycle, car, and bus — collected at tourism access gates in the Jember area and augmented using Roboflow. Six model variants were trained and evaluated: YOLOv8n, YOLOv8s, and YOLOv8m as baseline models, alongside their SPD-Conv-modified counterparts. The results show that the YOLOv8n baseline achieved the highest overall mAP@0.5 of 0.712, while among SPD-Conv models, YOLOv8s-SPD achieved the highest mAP@0.5 of 0.695 with the highest Precision of 0.641. The integration of SPD-Conv into the Small variant improved accuracy over its baseline (0.695 vs. 0.690) without increasing computational cost, as inference speed was even slightly faster (1.2 ms vs. 1.3 ms). The DeepSORT algorithm successfully assigned unique IDs to each detected vehicle, enabling accurate counting through a virtual counting line while preventing double-counting. This system demonstrates a promising approach for automated vehicle monitoring at tourism facilities.