ThesisOpen AccessMetadata saja
2025•Skripsi, Thesis, Disertasi, dan Tugas Akhir
Human error is still a major cause of accidents, including drivers' negligence in paying attention to and recognizing traffic signs on the road. Traffic signs serve to provide clear guidance for drivers, but it is often difficult for drivers to recognize these signs quickly, especially in non-ideal environmental conditions such as bad weather, low lighting, or unclear sign positions. This increases the risk of traffic accidents. This research proposes an automatic traffic sign detection system using YOLOv8 to recognize warning, prohibition, and command signs in real-time. The system is focused on recognizing various types of signs such as warnings, prohibitions, and commands in real-time in one image display, and the user interface is built with PyQt5 to display real-time detection results. The dataset used amounted to 2,210 images consisting of 780 primary datasets and 1,430 secondary datasets. After passing the annotation and augmentation stages, the dataset to be processed becomes 3,629 images with a division into 87% training data and 13% validation data. Testing the results of the training model shows excellent detection performance with a precision value of 99.8%, recall of 98.2%, and a mean average precision (mAP) value of 98.7% for IoU 50 89.7%.The system was also evaluated using 78 real-time images with various lighting conditions and angles.
ThesisOpen AccessMetadata saja
2025•
kali tidak konsisten terutama pada pengamatan berskala besar. Untuk mengatasi permasalahan tersebut, peneliti mengusulkan sistem deteksi tingkat kesegaran kubis putih secara otomatis dan real-time dengan menggunakan algoritma deep learning YOLOv8. Sistem ini akan mengidentifikasi berdasarkan tiga tingkatan kesegaran, yaitu kubis segar, kubis kurang segar, dan kubis busuk. Dataset yang digunakan terdiri dari 3009 citra kubis putih hasil dari proses pre-processing berupa resize, dan rotasi, serta proses augmentasi citra berupa brightness untuk meningkatkan variasi sampel sehingga diperoleh proporsi 93% (3009 citra) data latih, 3% (106 citra) data validasi, dan 3% (106 citra) data uji. Hasil pengujian sistem terhadap 106 citra data uji menghasilkan nilai rata rata recall sebesar 0,847, recall sebesar 0,878 dan f-measure sebesar 0,863. Kemudian hasil pengujian sistem secara real-time terhadap citra multi objek (berisi tiga kelas secara acak) yang diambil sebanyak 1 kali per 10 detik selama 5 menit pada jarak 20 cm dihasilkan nilai recall sebesar 0,919, recall sebesar 0,902, dan f-measure sebesar 0,903, sedangkan pada jarak 30 cm diperoleh nilai recall sebesar 0,889, recall sebesar 0,911, dan f-measure sebesar 0,888. Dari hasil pengujian tersebut membuktikan bahwa sistem dapat bekerja dalam mendeteksi kesegaran kubis putih secara real-time paling baik pada jarak 20 cm. Hal ini membuktikan bahwa usulan sistem dapat menjadi solusi efektif untuk seleksi mutu produk kubis putih secara otomatis dan real-time.
