Deteksi Lajur dan Identifikasi Marka Garis Jalan untuk Mobil Otonom Menggunakan Metode Semantic Segmentation Convolutional Neural Network

Rifian, Nabilla Ananda (2023) Deteksi Lajur dan Identifikasi Marka Garis Jalan untuk Mobil Otonom Menggunakan Metode Semantic Segmentation Convolutional Neural Network. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 07111940000161-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 September 2025. Download (2MB) | Request a copy

Abstract

Deteksi lajur merupakan salah satu hal yang krusial pada driving scene mobil otonom sebagai advanced driver assistance. Selain mendeteksi, mobil otonom perlu dapat membedakan arti marka garis yang umum berada di jalan raya, yaitu garis putus-putus dan utuh. Sejumlah metode deep learning dalam bidang computer vision telah dikembangkan seperti algoritma Hough Transform, dan dewasa ini telah dievaluasi model hybrid CNN-RNN oleh penelitian tugas akhir sebelumnya. Namun, hasil keduanya masih menyamakan jenis marka garis. Teknik semantic segmentation digunakan sebagai klasifikasi piksel menggunakan encoder-decoder CNN dengan menggunakan lapisan convolutional (FCN). Pada penelitian tugas akhir ini digunakan dua arsitektur yaitu SegNet dan UNet. Dalam pengambilan data untuk melatih model neural network, berbagai kondisi jalan yang diambil oleh kamera memberikan hasil yang berbeda-beda, tak jarang terdapat noise pada gambar dengan cahaya yang terdistorsi. Maka dari itu, noise tersebut perlu di filter dan filter yang digunakan adalah filter Wiener. Hasil pengujian diukur menggunakan nilai pixel accuracy dan mean IoU. Didapati rata-rata nilai pixel accuracy berturut-turut oleh model UNet dan SegNet tanpa dan dengan filter Wiener adalah 0.9976, 0.9972, 0.9978, dan 0.9964. Sementara rata-rata nilai mean IoU adalah 0.8934, 0.8808, 0.9024, dan 0.8499. Nilai tertinggi diperoleh UNet dengan menggunakan filter Wiener.
=================================================================================================================================
Lane detection is one crucial aspect of autonomous driving scenes as advanced driver assistance. Besides detection, autonomous vehicles need to distinguish the meaning of common lane markings on the road, namely dashed and solid lines. Several deep learning methods in the field of computer vision have been developed, such as the Hough Transform algorithm, and more recently, a hybrid CNN-RNN model has been evaluated by previous research studies. However, both approaches still treat the types of lane markings equally. The technique of semantic segmentation is used for pixel classification using an encoder-decoder CNN with convolutional layers (FCN). In this final research study, two architectures, namely SegNet and UNet, are utilized. During the data collection for training the neural network model, various road conditions captured by the camera yield different results, often introducing noise in the images with distorted lighting. Therefore, it is necessary to filter out this noise, and the filter used is the Wiener filter. The evaluation of the model performance is measured using pixel accuracy and mean IoU (Intersection over Union) values. The average pixel accuracy values obtained for UNet and SegNet, both with and without the Wiener filter, are 0.9976, 0.9972, 0.9978, and 0.9964, respectively. Meanwhile, the average mean IoU values are 0.8934, 0.8808, 0.9024, and 0.8499. The highest values are achieved by UNet when using the Wiener filter.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Mobil Otonom, Deteksi Lajur, Marka Garis, Deep Learning, Semantic Segmentation, Filter Wiener ============================================================ Autonomous Car, Lane Detection, Line Road Marking, Deep Learning, Semantic Segmentation, Wiener Filter
Subjects:	Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5102.9 Signal processing.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Rifian Nabilla Ananda
Date Deposited:	27 Jul 2023 06:35
Last Modified:	27 Jul 2023 06:35
URI:	http://repository.its.ac.id/id/eprint/99553

Actions (login required)

View Item