Perencanaan Gerakan Berbasis Perilaku Mobil Otonom Menggunakan Algoritma Deep Learning

Ar Rayan, Daffa Muhammad Adha (2024) Perencanaan Gerakan Berbasis Perilaku Mobil Otonom Menggunakan Algoritma Deep Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5022201127-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (2MB) | Request a copy

Abstract

Kendaraan otonom membutuhkan deteksi real-time terhadap lingkungan sekitar untuk memastikan navigasi yang aman. Penelitian ini menyelidiki integrasi algoritma deteksi objek dan deteksi jalur untuk pemahaman lingkungan yang komprehensif. YOLO (berbasis bounding box) memfasilitasi deteksi objek, sementara Ultra-Fast Lane Detection (berbasis segmentasi) menangani deteksi jalur. Pendekatan gabungan ini mengekstrak informasi penting, termasuk jarak objek, posisi jalur, dan kelengkungan jalur, sehingga memungkinkan mobil otonom membuat keputusan yang tepat. Tindakan yang dilakukan termasuk menghitung perlambatan dan mengeluarkan peringatan tabrakan, mengevaluasi perubahan jalur, dan mengantisipasi manuver yang akan datang. Analisis data menunjukkan keberhasilan integrasi YOLO dan Deteksi Jalur Sangat Cepat untuk klasifikasi multi-kelas. Kumpulan data COCO secara konsisten mengungguli kumpulan data COCO+Custom dalam pelatihan YOLO karena kualitasnya yang unggul. Deteksi objek menggunakan COCO menghasilkan akurasi yang tinggi, sementara COCO+Custom menghasilkan kesalahan deteksi. Dataset CULane menunjukkan kinerja terbaik dalam identifikasi jalan. Temuan ini menyoroti pentingnya dataset berkualitas tinggi untuk deteksi objek dan jalur. Lebih lanjut, penelitian ini meneliti ketergantungan tiga sistem peringatan perilaku mobil, khususnya Forward Collision Warning System (FCWS), Lane Keeping Assist System (LKAS), dan Lane Departure Warning System (LDWS). Ketiga sistem tersebut mengandalkan sistem deteksi lajur fungsional, bersama dengan elemen tambahan seperti deteksi objek, penghitungan jarak, penghitungan radius, dan penghitungan belokan. Penelitian di masa depan dapat mengeksplorasi set data yang lebih baik, metode anotasi alternatif, dan peningkatan iterasi pelatihan.
=====================================================================================================================================
Autonomous vehicles require real-time awareness of their surroundings to ensure safe navigation. This research investigates the integration of object detection and lane detection algorithms for comprehensive environmental understanding. YOLO (bounding box-based) facilitates object detection, while Ultra-Fast Lane Detection (segmentation-based) tackles lane detection. This combined approach extracts critical information, including object distance, lane position, and lane curvature, enabling the autonomous car to make informed decisions. Actions include calculating deceleration and issuing collision warnings, evaluating lane changes, and anticipating upcoming manuvers. Data analysis revealed successful integration of YOLO and Ultra-Fast Lane Detection for multi-class classification. The COCO dataset consistently outperformed the COCO+Custom dataset in YOLO training due to its superior quality. Object detection using COCO yielded high accuracy, while COCO+Custom resulted in misdetections. The CULane dataset demonstrated the best performance in road identification. These findings highlight the importance of high-quality datasets for both object and lane detection. Furthermore, the research examines the dependencies of three car behavior alert systems, specifically Forward Collision Warning System (FCWS), Lane Keeping Assist System (LKAS), and Lane Departure Warning System (LDWS). All three systems rely on a functional lane detection system, along with additional elements like object detection, distance calculation, radius calculation, and turning calculation. Future research could explore enhanced datasets, alternative annotation methods, and increased training iterations. Additionally, investigating the compatibility issues with the CurveLanes dataset and improving overall lane detection robustness are crucial areas for further development.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Mobil Otonom, Ultra-Fast Lane Detection, YOLO Autonomous vehicle, Ultra-Fast Lane Detection, YOLO
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T57.74 Linear programming
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Daffa Muhammad Adha Ar Rayan
Date Deposited:	22 Jul 2024 06:09
Last Modified:	22 Jul 2024 06:09
URI:	http://repository.its.ac.id/id/eprint/108628

Actions (login required)

View Item