DETEKSI JENIS KENDARAAN PADA JEMBATAN BERBASIS COMPUTER VISION DENGAN ALGORITMA YOLOv8

Saputra, Herry (2025) DETEKSI JENIS KENDARAAN PADA JEMBATAN BERBASIS COMPUTER VISION DENGAN ALGORITMA YOLOv8. Sarjana (S1) thesis, Universitas Islam 45.

	Text PENDAHULUAN.pdf Download (980kB)
	Text BAB I.pdf Restricted to Repository staff only Download (212kB) | Request a copy
	Text BAB II.pdf Restricted to Repository staff only Download (976kB) | Request a copy
	Text BAB III.pdf Restricted to Repository staff only Download (348kB) | Request a copy
	Text BAB IV.pdf Restricted to Repository staff only Download (446kB) | Request a copy
	Text BAB V.pdf Restricted to Repository staff only Download (155kB) | Request a copy
	Text DAFTAR PUSTAKA.pdf Download (162kB)

Abstract

Bridges are vital infrastructure that function as connectors between regions and are designed to withstand specific traffic loads. Given the limitations in load-bearing capacity and the risk of damage due to excessive loads, monitoring Average Daily Traffic (ADT) becomes crucial. High Average Daily Traffic (ADT), particularly from heavy vehicles, can accelerate structural degradation and increase the risk of premature damage. Additionally, bridges often become traffic bottlenecks, making accurate ADT data essential to support preventive maintenance planning and traffic flow management. This study employs a quantitative approach focusing on the performance evaluation of two object detection libraries (Library M and Library L) in identifying vehicles by type (cars, motorcycles, trucks) from video recordings. Test results show that Library L performs more reliably and consistently in terms of recall and F1 score, especially for trucks and cars, with recall values of 0.95 for cars and 0.67 for trucks. The F1 scores were 0.89 for cars and 0.65 for trucks, outperforming Library M in these metrics. However, Library M excels in precision, achieving the highest value of 0.85 for cars, and also in processing speed, with a time of 10.3 ms. Further testing with Library L showed the highest accuracy in the car category, where it detected 15 true positives (TP) out of 18 ground truth (GT) objects. In contrast, detection performance for motorcycles remained very low. These findings indicate that although Library L is generally more reliable, improvements are still needed in detecting two-wheeled vehicles and trucks to optimize the system for ADT monitoring on bridges.

Actions (login required)

View Item