Implementasi YOLOv8 dan TensorFlow Lite pada Aplikasi Mobile untuk Deteksi Kerusakan Jalan secara Real-Time
DOI:
https://doi.org/10.36312/qwq2bk22Abstract
Kerusakan jalan seperti lubang dan retak memerlukan mekanisme pelaporan yang cepat, objektif, dan berbasis lokasi agar proses pemeliharaan infrastruktur dapat dilakukan secara lebih responsif. Penelitian ini bertujuan mengembangkan aplikasi mobile berbasis computer vision untuk mendeteksi dan melaporkan kerusakan jalan secara real-time dengan mengintegrasikan model YOLOv8, TensorFlow Lite, kamera smartphone, GPS, basis data lokal, RESTful API Laravel, MySQL, dan dashboard monitoring geografis. Metode penelitian menggunakan pendekatan penelitian dan pengembangan perangkat lunak dengan model Waterfall yang mencakup analisis kebutuhan, desain sistem, implementasi, pengujian fungsional, dan evaluasi model deteksi. Dataset disusun secara hibrida dari dataset publik dan data lokal, lalu dibagi menjadi data latih 70%, validasi 20%, dan uji 10%. Hasil menunjukkan bahwa aplikasi berhasil menjalankan autentikasi, deteksi kamera, pengambilan koordinat, pengiriman laporan daring, penyimpanan laporan saat blank spot, visualisasi peta GIS, dan validasi administrator. Model YOLOv8s yang dikonversi ke format TensorFlow Lite memperoleh rata-rata precision 69,4%, recall 66,6%, dan mAP50 67,7%. Kinerja terbaik terdapat pada kelas pothole dengan mAP50 76,6%, sedangkan kelas crack memperoleh mAP50 58,8% karena bentuk retakan yang tipis dan tidak beraturan. Temuan ini menunjukkan bahwa integrasi edge computing dan server terpusat dapat menjadi solusi awal untuk pelaporan kerusakan jalan yang lebih cepat, meskipun peningkatan dataset lokal dan pendekatan segmentasi masih diperlukan untuk memperbaiki deteksi retakan.
Road damage, including potholes and cracks, requires a fast, objective, and location-based reporting mechanism to support more responsive infrastructure maintenance. This study developed a mobile computer-vision application for real-time road damage detection and reporting by integrating YOLOv8, TensorFlow Lite, smartphone cameras, GPS, local storage, a Laravel RESTful API, MySQL, and a geographic monitoring dashboard. The research adopted a software research-and-development approach using the Waterfall model, covering requirement analysis, system design, implementation, functional testing, and detection-model evaluation. The image dataset was prepared using a hybrid approach that combined public datasets and local field data, with a 70% training, 20% validation, and 10% testing split. The results show that the application successfully performed authentication, camera-based detection, coordinate acquisition, online report submission, offline report storage in blank-spot conditions, GIS map visualization, and administrator validation. The YOLOv8s model converted into TensorFlow Lite achieved an average precision of 69.4%, recall of 66.6%, and mAP50 of 67.7%. The best performance was obtained for pothole detection with an mAP50 of 76.6%, while crack detection reached an mAP50 of 58.8% because cracks are thin, elongated, and irregular visual objects. These findings indicate that the integration of edge computing and centralized server management can provide an initial solution for faster road damage reporting, although larger local datasets and segmentation-based approaches are still needed to improve crack detection.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Syauqi Revo Mardian, Khairi Budayawan, Titi Sriwahyuni, Hadi kurnia

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
Authors who publish with Journal of Authentic Research agree to the following terms:
- For all articles published in Journal of Authentic Research, copyright is retained by the authors. Authors give permission to the publisher to announce the work with conditions. When the manuscript is accepted for publication, the authors agrees to implement a non-exclusive transfer of publishing rights to the journals.
- Authors retain copyright and grant the journal right of first publication with the work simultaneously licensed under a Creative Commons Attribution-ShareAlike 4.0 International License that allows others to share the work with an acknowledgment of the work's authorship and initial publication in this journal.
- Authors are able to enter into separate, additional contractual arrangements for the non-exclusive distribution of the journal's published version of the work (e.g., post it to an institutional repository or publish it in a book), with an acknowledgment of its initial publication in this journal.
- Authors are permitted and encouraged to post their work online (e.g., in institutional repositories or on their website) prior to and during the submission process, as it can lead to productive exchanges, as well as earlier and greater citation of published work (See The Effect of Open Access).

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.