Skip to content
Open access

OPTIMIZATION OF A YOLOV5 AND KALMAN FILTER-BASED VEHICLE TRACKING SYSTEM WITH VEHICLE MOVEMENT DIRECTION ANALYSIS IN CCTV VIDEO

Sep 2026 · JURTEKSI (Jurnal Teknologi dan Sistem Informasi) · 0 citations

Abstract

Abstract: This study aims to optimize a video-based vehicle tracking system by integrating YOLOv5, the Kalman Filter, and vehicle movement direction analysis to improve tracking stability under real-world traffic conditions. The dataset consists of four traffic CCTV videos collected from four CCTV cameras at the signalized intersection of Dr. Wahidin Street, R.A. Kartini Street, and Ade Irma Suryani Street in Pematangsiantar City, Indonesia, which were preprocessed into image frames. YOLOv5 was employed for object detection, while the Kalman Filter was used to maintain tracking continuity during occlusion and temporary detection loss. Experimental results show that YOLOv8n achieved better detection performance with 100% accuracy, 100% recall, and 95% precision, whereas YOLOv5nu obtained 100% accuracy, 67% recall, and 89% precision. Nevertheless, YOLOv5nu was selected as the primary model due to its higher computational efficiency for real-time applications. The proposed system achieved a MOTA of 95.73%, MOTP of 82.61%, and an IDF1 score of 94.28%, demonstrating stable and accurate vehicle tracking performance.             Keywords: CCTV video; direction analysis; kalman filter; vehicle tracking; YOLOv5.                                                                                                                       Abstrak: Penelitian ini bertujuan mengoptimalkan sistem pelacakan kendaraan berbasis video dengan mengintegrasikan YOLOv5, Kalman Filter, dan analisis arah pergerakan kendaraan untuk meningkatkan stabilitas pelacakan pada kondisi lalu lintas nyata. Dataset yang digunakan terdiri dari empat video CCTV lalu lintas yang diperoleh dari empat kamera CCTV di persimpangan Jalan Dr. Wahidin, Jalan R.A. Kartini, dan Jalan Ade Irma Suryani, Kota Pematangsiantar, kemudian diproses menjadi frame melalui tahap preprocessing. YOLOv5 digunakan sebagai detektor objek, sedangkan Kalman Filter diterapkan untuk menjaga kontinuitas pelacakan pada kondisi occlusion dan kehilangan deteksi sementara. Hasil pengujian menunjukkan bahwa YOLOv8n memperoleh performa deteksi lebih baik dengan akurasi 100%, recall 100%, dan precision 95%, sedangkan YOLOv5nu memperoleh akurasi 100%, recall 67%, dan precision 89%. Meskipun demikian, YOLOv5nu dipilih sebagai model utama karena lebih efisien untuk aplikasi real-time. Sistem yang diusulkan juga menghasilkan MOTA 95,73%, MOTP 82,61%, dan IDF1 94,28%, yang menunjukkan kemampuan pelacakan kendaraan yang stabil dan akurat.   Kata kunci: analisis arah; kalman filter; pelacakan kendaraan; video CCTV; YOLOv.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.