Simulasi Autonomous Robot Menggunakan Nvidia Jetbot Dalam Klasifikasi Lampu Lalu Lintas

Baskoro, Bonifasius Jeremy (2022) Simulasi Autonomous Robot Menggunakan Nvidia Jetbot Dalam Klasifikasi Lampu Lalu Lintas. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 02111640000104-Undergraduate_Thesis.pdf Download (5MB)

Abstract

Penggunaan autonomous robot untuk membantu pekerjaan manusia sudah mulai dilakukan. Mulai dari industri manufaktur, industri logistic, maupun industri medis. Tujuan dari penggunaan autonomous robot adalah untuk memudahkan manusia serta melakukan kegiatan pekerjaan lebih efektif. Autonomous robot dilengkapi dengan kamera untuk mengetahui lingkungan sekitarnya. Masukan dari kamera diproses oleh komputer menggunakan beberapa metode seperti computer vision, machine learning, deep learning. Salah satu alat yang dapat digunakan untuk menerapkan autonomous robot adalah NVIDIA Jetbot. Dengan tujuan agar autonomous robot dapat berinteraksi dengan baik dengan lampu lalu lintas maka dilakukan engaplikasian menggunakan NVIDIA Jetbot. Salah satu metode untuk NVIDIA Jetbot dapat berfungsi sebagai autonomous robot delivery adalah dengan dapat mengklasifikasi lampu lalu lintas seperti skenario umumnya pada jalanan di Indonesia. Pada penelitian ini digunakan ResNet karena arsitektur ResNet dinilai mempunyai nilai akurasi yang tinggi serta dengan model yang mempunyai parameter yang sedikit. NVIDIA Jetbot akan menjalankan fungsi klasifikasi lampu lalu lintas dengan model ResNet dan input dari kamera Jetbot. Pengujian klasifikasi lampu lalu lintas dilakukan pada 4 tingkat kecepatan yang berbeda, 2 variasi kondisi pencahayaan lintasan, dan serta dataset training yang berbeda yaitu terdiri dari 500 gambar Bosch Small Traffic Light Dataset dengan augmentasi dan tanpa augmentasi, serta dataset berupa gambar diambil dari kamera Jetbot dengan jumlah 250 dan 500 gambar. Hasil Performa Jetbot dalam klasifikasi lampu lalu lintas pada pengujian di kondisi pencahayaan terang dengan kecepatan berturut-turut sebesar 0,3, 0,4, 0,6, dan 0,8 untuk data training 500 gambar Bosch Small Traffic Light Dataset tanpa augmentasi yaitu 50,67%, 40,67%, 34,67%, 29,33%. Untuk data training 500 Gambar Bosch Small Traffic Light Dataset dengan augmentasi yaitu 58%, 55,33%, 44,67%, dan 30,67%. Untuk data training 250 gambar diambil dari kamera Jetbot yaitu 93,33%, 92%, 82,67%, dan 70,67%. Untuk data training 500 gambar diambil dari kamera Jetbot yaitu 96%, 94,67%, 86,67%, dan 72,67%. Pada kondisi pencahayaan gelap dengan kecepatan berturut-berturut sebesar sebesar 0,3, 0,4, 0,6, dan 0,8 untuk data training 500 gambar Bosch Small Traffic Light Dataset tanpa augmentasi yaitu 40,67%, 39,33%, 32,67%, dan 28,00%. Untuk data training 500 Gambar Bosch Small Traffic Light Dataset dengan augmentasi yaitu 59,33%, 56,67%, 53,33%, dan 29,33%. Untuk data training 250 gambar diambil dari kamera Jetbot yaitu 93,33%, 94%, 88,67%, dan 68%. Untuk data training 500 gambar diambil dari kamera Jetbot yaitu 97,33%, 94%, 84,67%, dan 82%.
=================================================================================================================================
The use of autonomous robots to help human work has begun. Starting from the manufacturing industry, the logistics industry, and the medical industry. The purpose of using autonomous robots is to make it easier for humans to carry out work activities more effectively. Autonomous robot is equipped with a camera to know the surrounding environment. Input from the camera is processed by a computer using several methods such as computer vision, machine learning, deep learning. One of the tools that can be used to implement autonomous robots is NVIDIA Jetbot. With the aim that autonomous robots can interact well with traffic lights, an application is made using NVIDIA Jetbot. One of the methods for the NVIDIA Jetbot to function as an autonomous robot delivery is to be able to classify traffic lights as in general scenarios on roads in Indonesia. In this study, ResNet is used because the ResNet architecture is considered to have a high accuracy value and with a model that has few parameters. NVIDIA Jetbot will perform the traffic light classification function with the ResNet model and input from the Jetbot camera. Traffic light classification tests were carried out at 4 different speed levels, 2 variations of track lighting conditions, and a different training dataset consisting of 500 images of the Bosch Small Traffic Light Dataset with and without augmentation, as well as a dataset in the form of images taken from a Jetbot camera with the number of 250 and 500 images. Jetbot performance results in traffic light classification on tests in bright lighting conditions with successive speeds of 0.3, 0.4, 0.6, and 0.8 for training data of 500 images of Bosch Small Traffic Light Dataset without augmentation, which is 50 ,67%, 40.67%, 34.67%, 29.33%. For training data 500 Images of Bosch Small Traffic Light Dataset with augmentation are 58%, 55.33%, 44.67%, and 30.67%. For training data, 250 images were taken from the Jetbot camera, namely 93.33%, 92%, 82.67%, and 70.67%. For training data 500 images were taken from the Jetbot camera, namely 96%, 94.67%, 86.67%, and 72.67%. In dark lighting conditions with successive speeds of 0.3, 0.4, 0.6, and 0.8 for training data of 500 images of Bosch Small Traffic Light Dataset without augmentation, namely 40.67%, 39.33%, 32.67%, and 28.00%. For training data 500 Images of Bosch Small Traffic Light Dataset with augmentation are 59.33%, 56.67%, 53.33%, and 29.33%. For training data, 250 images were taken from the Jetbot camera, namely 93.33%, 94%, 88.67%, and 68%. For training data 500 images were taken from the Jetbot camera, namely 97.33%, 94%, 84.67%, and 82%.

Item Type:	Thesis (Other)
Additional Information:	RSM 629.892 Bas s-1 2022
Uncontrolled Keywords:	Autonomous Robot, Machine Learning, Deep Learning, NVIDIA Jetbot
Subjects:	T Technology > TJ Mechanical engineering and machinery > TJ211.415 Mobile robots
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Mechanical Engineering > 21201-(S1) Undergraduate Thesis
Depositing User:	Mr. Marsudiyana -
Date Deposited:	13 Feb 2025 03:52
Last Modified:	13 Feb 2025 03:52
URI:	http://repository.its.ac.id/id/eprint/118711

Actions (login required)

View Item