Sistem Verifikasi Otomatis Pembacaan Odometer Sepeda Motor Berbasis Yolo, Cnn, dan Dinov2

Richie, Junathan (2026) Sistem Verifikasi Otomatis Pembacaan Odometer Sepeda Motor Berbasis Yolo, Cnn, dan Dinov2. Project Report. [s.n.]. (Unpublished)

Text 5025231019-Project_Report.pdf - Accepted Version Restricted to Repository staff only Download (1MB) | Request a copy

Abstract

Pencatatan nilai mileage (odometer) pada foto speedometer sepeda motor saat ini masih diverifikasi secara manual oleh petugas, sehingga rawan kesalahan baca maupun manipulasi foto pada proses klaim servis maupun garansi. Kerja praktik ini membahas perancangan dan implementasi sistem verifikasi otomatis pembacaan odometer berbasis kecerdasan buatan, yang dikembangkan di Laboratorium Algoritma dan Pemrograman bekerja sama dengan PT XYZ. Sistem berupa backend FastAPI yang menggabungkan PaddleOCR untuk koreksi orientasi gambar dan pengenalan pola tripmeter, YOLOv11 untuk deteksi dan pemotongan area speedometer sekaligus deteksi bounding box tiap digit, sebuah CNN digit classifier untuk mengklasifikasi nilai tiap digit, serta DinoV2 dan basis data vektor Qdrant untuk memvalidasi kesesuaian tipe mesin pada foto dengan data yang diklaim pengguna. Pengujian pembacaan mileage terbaru terhadap 554 foto odometer menunjukkan akurasi exact-match 89,53% pada konfigurasi produksi. Hasil evaluasi menunjukkan sistem mampu menolak foto tripmeter, foto buram, dan foto tipe mesin yang tidak sesuai, serta memberikan tingkat kepercayaan (confidence) pada setiap prediksi. Hasil kerja praktik ini diharapkan dapat menjadi dasar pengembangan sistem verifikasi odometer yang lebih akurat dan dapat diandalkan untuk mendukung proses bisnis PT XYZ.
=================================================================================================================================
Recording mileage (odometer) values from motorcycle speedometer photos is currently still verified manually by staff, making the process prone to reading errors and photo manipulation during service or warranty claim procedures. This internship report discusses the design and implementation of an artificial intelligence-based automatic odometer reading verification system, developed at the Algorithm and Programming Laboratory in collaboration with PT XYZ. The system consists of a FastAPI backend that integrates PaddleOCR for image orientation correction and tripmeter pattern recognition, YOLOv11 for detecting and cropping the speedometer area as well as detecting bounding boxes for each digit, a CNN digit classifier to classify the value of each digit, and DinoV2 together with the Qdrant vector database to validate whether the engine type in the photo matches the data claimed by the user. The latest mileage-reading test on 554 odometer photos achieved an exact-match accuracy of 89.53% under the production configuration. Evaluation results show that the system is able to reject tripmeter photos, blurry photos, and photos of mismatched engine types, while also providing a confidence level for each prediction. The results of this internship are expected to serve as a foundation for developing a more accurate and reliable odometer verification system to support PT XYZ's business processes.

Item Type:	Monograph (Project Report)
Uncontrolled Keywords:	odometer, YOLO, CNN, DinoV2, image verification, verifikasi citra.
Subjects:	Q Science > QA Mathematics > QA76.6 Computer programming. T Technology > T Technology (General)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Junathan Richie
Date Deposited:	17 Jul 2026 07:52
Last Modified:	17 Jul 2026 07:52
URI:	http://repository.its.ac.id/id/eprint/135243

Actions (login required)

View Item