Penelitian ini bertujuan untuk mengembangkan dan mengevaluasi sistem deteksi kendaraan menggunakan algoritma You Only Look Once versi 8 (YOLOv8) pada rekaman video lalu lintas di Sleman, Yogyakarta. Dataset yang digunakan berasal dari rekaman CCTV yang mencakup berbagai jenis kendaraan dengan kondisi pencahayaan dan kepadatan lalu lintas yang beragam. Proses preprocessing meliputi anotasi objek, normalisasi, dan augmentasi citra untuk meningkatkan kualitas serta keragaman data. Model dilatih menggunakan konfigurasi parameter yang dioptimalkan dalam kerangka deep learning. Evaluasi kinerja dilakukan berdasarkan precision, recall, F1-score, mean average precision (mAP), serta kecepatan deteksi dalam frame per second (FPS). Hasil penelitian menunjukkan bahwa YOLOv8 mampu mendeteksi motor, mobil, truk, dan bus secara efektif dengan nilai precision sebesar 0,869, recall 0,831, F1-score 0,849, mAP50 0,897, serta kecepatan rata-rata 28 FPS. Hasil uji manual memberikan precision 0,99, recall 0,78, F1-score 0,87, dan akurasi keseluruhan 77,2%. Kendala utama ditemukan pada kondisi pencahayaan rendah dan objek yang saling menutupi, khususnya untuk kendaraan berukuran kecil. Secara keseluruhan, penelitian ini membuktikan bahwa YOLOv8 berpotensi mendukung pengembangan sistem pemantauan lalu lintas yang responsif dan efisien di lingkungan nyata.

This study aims to develop and evaluate a vehicle detection system using the You Only Look Once version 8 (YOLOv8) algorithm on traffic video recordings in Sleman, Yogyakarta. The dataset was obtained from CCTV recordings covering various vehicle types under diverse lighting and traffic density conditions. The preprocessing stage included object annotation, normalization, and image augmentation to improve data quality and diversity. The model was trained with optimized parameter configurations within a deep learning framework. Performance evaluation was carried out using precision, recall, F1-score, mean average precision (mAP), and detection speed measured in frames per second (FPS). The results show that YOLOv8 effectively detects motorcycles, cars, trucks, and buses with precision of 0.869, recall of 0.831, F1-score of 0.849, mAP50 of 0.897, and an average speed of 28 FPS. Manual evaluation yielded precision of 0.99, recall of 0.78, F1-score of 0.87, and an overall accuracy of 77.2%. The main challenges occurred under low lighting and occlusion conditions, especially for small vehicles. Overall, this study demonstrates that YOLOv8 has strong potential to support the development of responsive and efficient traffic monitoring systems in real-world environments.