Optimalisasi Sistem Deteksi Rekomendasi Persimpangan Jalan dalam Kepadatan Lalu Lintas Menggunakan Algoritma YOLO v11

Nurdiansyah, Aziz
Kalifia, Anna Dina

TIN: TERAPAN INFORMATIKA NUSANTARA
Vol 6 No 7 (2025): December 2025

The crucial problem facing modern urban areas is traffic congestion, which causes significant time, economic, and environmental losses. Manual identification of traffic density has proven to be inefficient and error-prone, especially at intersections with high real-time density fluctuations. This research aims to design and test a real-time vehicle detection system that is robust against environmental variability and capable of providing accurate predictions of density levels, contributing to the transformation of traffic management from reactive to predictive.As a solution, a traffic density detection and analysis system based on Deep Learning is proposed, utilizing an optimized YOLOv11 model, integrating Image Processing and a Neural Network. YOLOv11 is used to accurately detect and classify various types of vehicles from CCTV video footage, even in low-light conditions, and the results serve as input for the Adaptive Traffic Light Control Module based on the Density.Preliminary results from model training show very fast convergence, achieving a comprehensive accuracy (mAP@0.5) of 0.956 on the validation set in just 10 epochs. Although testing on new test data yielded an overall class mAP@0.5 of 0.631, the model demonstrated superior performance for detecting large vehicles, such as trucks (mAP@0.5 = 0.962) and cars (mAP@0.5 = 0.935). This system is expected to provide accurate traffic density information, enable adaptive traffic light settings, and ultimately contribute to intelligent traffic management systems.