Sasmito, Bandi (2026) MODEL SISTEM TERPADU DAN TEPAT GUNA UNTUK PENGUKURAN KERUSAKAN PERKERASAN JALAN BERBASIS PENGINDRAAN JAUH DAN POSITIONING. Doctoral thesis, UNDIP.

	Text Cover.pdf Download (864kB)
	Text BAB 1.pdf Restricted to Repository staff only Download (642kB)
	Text BAB 2.pdf Restricted to Repository staff only Download (2MB)
	Text BAB 3.pdf Restricted to Repository staff only Download (1MB)
	Text BAB 4.pdf Restricted to Repository staff only Download (1MB)
	Text BAB 5.pdf Restricted to Repository staff only Download (1MB)
	Text BAB 6.pdf Restricted to Repository staff only Download (228kB)
	Text Daftar Pustaka.pdf Restricted to Repository staff only Download (365kB)

Abstract

Roads are important as a support for activities in human life in many interests. Its role is
urgently needed in facilitating access to transportation for humans, goods, and even
animals. Roads are one of the important infrastructures in land transportation for the
community. Activities of transporting people and goods on land in terms of economy and
non-economy by road. Behind all the advantages of the road is the risk of road damage.
Monitoring the condition of the road from damage is important so that it can be repaired
immediately. The implementation of monitoring still uses visual methods that have
inconsistent and slow weaknesses. So, this research aims to create a system of equipment
and applications to detect road damage based on artificial intelligence. The system
consists of a low-cost device configuration with camera sensors and a Global Navigation
System (GNSS) module for recording with an Android-based application, a deep learning
application to recognize road damage with YOLOv8 (You Only Look Once), and a
calculation of the extent of road damage detected. The method uses applied research,
namely, applying sensor technology and artificial intelligence to detect road damage.
Camera sensors and GNSS module for image and video recording are connected and
controlled with an Android-based app installed on a smartphone. The camera sensor can
use the built-in smartphone or external camera connected with a USB (Universal Serial
Bus) cable. Deep learning applications with the YOLOv8 model are built in a desktop
format with an interface using the Python Tkinter library. The selection of YOLOv8 is
based on the model's stability and its ability to generalize to objects with high variations
in shape, size, and image conditions, making it more efficient and reliable for field
implementation based on image and video data with limited resources. The YOLOv8
model dataset is built from Global Road Damage Detection Challenge (GRDDC) data
and self-acquired results. The results achieved from this study are a system of equipment
for the acquisition of road data in the form of images and videos and an application for
identifying the location of the distribution of damage to the road contained in the image,
along with the position of coordinates that are accurate and easy to find back to the field.
The model demonstrated a satisfactory overall performance, achieving a mean Average
Precision (mAP) of 0.84 (84.21%). The average Intersection over Union (IoU) value
obtained from the detection samples was 0.871, indicating that the model achieved an
87% accuracy in delineating object boundaries through bounding boxes. Field
verification results further confirmed a strong correspondence between the model outputs
and the ground-truth data, with planimetric discrepancies ranging from 0.670 m to 9.138
m, an average deviation of 4.937 m, and a root mean square (RMS) error of 5.523 m.
Keywords: Road Damage Detection, Artificial Intelligence, Deep Learning, YOLO (You
Only Look Once), YOLOv8, Global Navigation Satelite System (GNSS)

Actions (login required)

View Item