Pengembangan Sistem Pendeteksi Penggunaan Sabuk Pengaman Menggunakan Convolutional Neural Network (CNN)

Cahyadi, Fahmi Fidyan (2020) Pengembangan Sistem Pendeteksi Penggunaan Sabuk Pengaman Menggunakan Convolutional Neural Network (CNN). Undergraduate thesis, InstitutTeknologi Sepuluh Nopember.

[img] Text
07211540000033-Undergraduate_Thesis.pdf - Accepted Version
Restricted to Repository staff only

Download (8MB) | Request a copy

Abstract

Salah satu penerapan computer vision ialah pendeteksian penggunaan sabuk pengaman. Banyak penelitian yang mengusulkan berbagai solusi berbasis pengolahan citra untuk mendeteksi penggunaan sabuk pengaman. Oleh karena itu dengan memanfaatkan Convolutional Neural Network, dibuatlah sebuah sistem yang dapat mendeteksi penggunaan sabuk pengaman penumpang kursi depan secara otomatis yang akan diterapkan di jalan raya. Proses training dilakukan menggunakan YOLOv3-SPP dan YOLOv3-tiny dengan menggunakan 1240 dataset. Data tersebut dibagi menjadi 969 image train dan 271 image test. Hasil training tertinggi yang diperoleh YOLOv3-SPP sebesar 90.2% mAP dan Hasil training tertinggi yang diperoleh YOLOv3-tiny sebesar 92% mAP. Hasil pengujian performa didapatkan mean Average Precision (mAP) tertinggi pada YOLOv3-SPP dengan nilai 90,2% dibandingkan dengan YOLOv3tiny yang mendapatkan nilai 92,6%. Dalam waktu pemrosesan dataset test, YOLOv3-tiny memiliki keunggulan waktu proses untuk mendeteksi 271 dataset, dibutuhkan waktu 7 detik, sementara pada YOLOv3-SPP, dibutuhkan waktu deteksi selama 15 detik untuk seluruh gambar. Kesalahan dalam deteksi penggunaan sabuk pengaman yaitu paling rendah hanya 5%. ========================================================= One application of image processing is the detection of the use of seat belts. Many studies have proposed various image processing based solutions to detect the use of seat belts. Therefore, by utilizing the Convolutional Neural Network, a system that can detect the use of front seat passenger belts is automatically created which will be applied on the highway. The training process is done using YOLOv3-SPP and YOLOv3-tiny using 1240 datasets. The data is divided into 969 image train and 271 image tests. The highest training results obtained by YOLOv3-SPP were 90.2% mAP and the highest training results obtained by YOLOv3-tiny were 92% mAP. The performance test results obtained the highest average precision (mAP) on YOLOv3-SPP with a value of 90.2% compared to YOLOv3tiny which scored 92.6%. In the processing time for dataset tests, YOLOv3-tiny has the advantage of processing time to detect 271 datasets, it takes 7 seconds, while in YOLOv3-SPP, it takes 15 seconds for all images. Errors in detection of the use of seat belts is at the lowest of only 5%.

Item Type: Thesis (Undergraduate)
Uncontrolled Keywords: Deteksi, Seat Belt Violation Detection, You Only Look Once (YOLO), Convolutional Neural Network (CNN). Detection, Seat Belt Violation Detection, You Only Look Once (YOLO), Convolutional Neural Network (CNN).
Subjects: T Technology > T Technology (General) > T57.83 Dynamic programming
T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques
Divisions: Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Computer Engineering > 90243-(S1) Undergraduate Thesis
Depositing User: Fahmi Fidyan Cahyadi
Date Deposited: 21 Aug 2020 03:25
Last Modified: 21 Aug 2020 03:25
URI: http://repository.its.ac.id/id/eprint/78926

Actions (login required)

View Item View Item