PRASETYO, NUGROHO ADI (2026) DETEKSI DAN PELACAKAN KENDARAAN MENGGUNAKAN YOLOv8 DAN DeepSORT DENGAN OPTIMASI METODE TAGUCHI. S1 thesis, Universitas Mercu Buana Jakarta.
|
Text (HAL COVER)
01 Cover.pdf Download (340kB) | Preview |
|
![[img]](https://repository.mercubuana.ac.id/style/images/fileicons/text.png) |
Text (BAB I)
02 Bab 1.pdf Restricted to Registered users only Download (106kB) |
|
|
Text (BAB II)
03 Bab 2.pdf Restricted to Registered users only Download (447kB) |
|
|
Text (BAB III)
04 Bab 3.pdf Restricted to Registered users only Download (629kB) |
|
|
Text (BAB IV)
05 Bab 4.pdf Restricted to Registered users only Download (966kB) |
|
|
Text (BAB V)
06 Bab 5.pdf Restricted to Registered users only Download (98kB) |
|
|
Text (DAFTAR PUSTAKA)
07 Daftar Pustaka.pdf Restricted to Registered users only Download (153kB) |
|
|
Text (LAMPIRAN)
08 Lampiran.pdf Restricted to Registered users only Download (4MB) |
Abstract
Automatic vehicle detection and tracking in video is an important component in developing Intelligent Transportation Systems (ITS), particularly for real-time traffic monitoring. This study aims to build a video-based vehicle detection and tracking system using YOLOv8m and DeepSORT, while applying hyperparameter optimization through Grid Search, Random Search, and the Taguchi Method, and comparing their performance. The variables studied include four vehicle object classes, namely motorcycle, car, truck, and bus, with optimized hyperparameters consisting of batch size, learning rate, and optimizer, and evaluation metrics comprising Precision, Recall, F1-score, mAP50, and mAP50-95. This research uses a quantitative experimental approach with a sample of 16,166 vehicle images, consisting of 11,536 training data, 2,533 validation data, and 2,097 testing data. The data were collected by combining public datasets from Roboflow Universe and Kaggle with independently collected traffic video recordings, which were then selected, cleaned, and standardized into YOLO annotation format. The Taguchi Method was implemented using an Orthogonal Array L4 with three factors at two levels, requiring only four experiments. The results show that the baseline model achieved an mAP50 of 0.8714, which increased to 0.9094 through Grid Search and 0.8982 through Random Search. The Taguchi Method delivered the best performance with an mAP50 of 0.9154 and mAP50-95 of 0.8069, increases of 5.05% and 4.53% over the baseline, using only four trials versus six for Random Search. The best Taguchi-optimized model was then integrated with DeepSORT and maintained consistent vehicle identities across frames under normal traffic conditions. It can be concluded that the Taguchi Method is the most effective and efficient hyperparameter optimization approach for the YOLOv8m model compared to Grid Search and Random Search, making it suitable for adoption in ITS-based vehicle detection and tracking systems. Keywords: vehicle detection, object tracking, YOLOv8, DeepSORT, Taguchi method. Deteksi dan pelacakan kendaraan secara otomatis pada video merupakan komponen penting dalam pengembangan sistem transportasi cerdas (Intelligent Transportation System/ITS), khususnya untuk mendukung pemantauan lalu lintas secara real-time. Penelitian ini bertujuan membangun sistem deteksi dan pelacakan kendaraan berbasis video menggunakan YOLOv8m dan DeepSORT, serta menerapkan optimasi hyperparameter menggunakan Grid Search, Random Search, dan Metode Taguchi untuk selanjutnya dibandingkan performanya. Variabel yang diteliti meliputi empat kelas objek kendaraan, yaitu motorcycle, car, truck, dan bus, dengan hyperparameter yang dioptimasi berupa batch size, learning rate, dan optimizer, serta metrik evaluasi berupa Precision, Recall, F1-score, mAP50, dan mAP50-95. Penelitian menggunakan pendekatan kuantitatif eksperimental dengan sampel sebanyak 16.166 citra kendaraan, terdiri atas 11.536 data pelatihan, 2.533 data validasi, dan 2.097 data pengujian. Data dihimpun dengan menggabungkan dataset publik dari Roboflow Universe dan Kaggle bersama data hasil pengambilan mandiri melalui rekaman video lalu lintas, yang kemudian diseleksi, dinormalisasi, dan diseragamkan ke dalam format anotasi YOLO. Metode Taguchi diterapkan menggunakan Orthogonal Array L4 dengan tiga faktor pada dua level sehingga hanya membutuhkan empat eksperimen. Hasil penelitian menunjukkan model baseline memperoleh mAP50 sebesar 0,8714, meningkat menjadi 0,9094 melalui Grid Search dan 0,8982 melalui Random Search. Metode Taguchi memberikan performa terbaik dengan mAP50 sebesar 0,9154 dan mAP50-95 sebesar 0,8069, atau meningkat 5,05% dan 4,53% dibandingkan baseline, meskipun hanya menggunakan empat percobaan dibandingkan enam percobaan pada Random Search. Model terbaik hasil Taguchi selanjutnya diintegrasikan dengan DeepSORT dan mampu mempertahankan identitas kendaraan secara konsisten antar-frame pada kondisi lalu lintas normal. Dapat disimpulkan bahwa Metode Taguchi merupakan pendekatan optimasi hyperparameter yang paling efektif dan efisien untuk model YOLOv8m dibandingkan Grid Search dan Random Search, sehingga layak diadopsi dalam pengembangan sistem deteksi dan pelacakan kendaraan berbasis ITS. Kata Kunci: deteksi kendaraan, pelacakan objek, YOLOv8, DeepSORT, metode Taguchi
| Item Type: | Thesis (S1) |
|---|---|
| NIM/NIDN Creators: | 41523110085 |
| Uncontrolled Keywords: | deteksi kendaraan, pelacakan objek, YOLOv8, DeepSORT, metode Taguchi |
| Subjects: | 000 Computer Science, Information and General Works/Ilmu Komputer, Informasi, dan Karya Umum > 000. Computer Science, Information and General Works/Ilmu Komputer, Informasi, dan Karya Umum > 004 Data Processing, Computer Science/Pemrosesan Data, Ilmu Komputer, Teknik Informatika 300 Social Science/Ilmu-ilmu Sosial > 380 Commerce, Communications, Transportation (Perdagangan, Komunikasi, Transportasi) > 388 Ground Transportation/Transportasi Jalan Raya > 388.3 Vehicular Transportation/Transportasi kendaraan |
| Divisions: | Fakultas Ilmu Komputer > Informatika |
| Depositing User: | khalimah |
| Date Deposited: | 16 Sep 2026 02:55 |
| Last Modified: | 16 Sep 2026 02:55 |
| URI: | http://repository.mercubuana.ac.id/id/eprint/103911 |
Actions (login required)
 |
View Item |
