Alif Ilman Nafian, - (2024) DETEKSI OBJEK BERBASIS LIDAR DENGAN METODE POINTPILLARS UNTUK KENDARAAN LISTRIK OTONOM PADA LINGKUNGAN TERBATAS. S1 thesis, Universitas Pendidikan Indonesia.
![]() |
Text
S_RPL_2009431_Title.pdf Download (428kB) |
![]() |
Text
S_RPL_2009431_Chapter1.pdf Download (269kB) |
![]() |
Text
S_RPL_2009431_Chapter2.pdf Restricted to Staf Perpustakaan Download (2MB) |
![]() |
Text
S_RPL_2009431_Chapter3.pdf Download (620kB) |
![]() |
Text
S_RPL_2009431_Chapter4.pdf Restricted to Staf Perpustakaan Download (670kB) |
![]() |
Text
S_RPL_2009431_Chapter5.pdf Download (217kB) |
![]() |
Text
S_RPL_2009431_Appendix.pdf Restricted to Staf Perpustakaan Download (8MB) |
Abstract
Pengembangan sistem deteksi objek untuk kendaraan listrik otonom menggunakan metode lokalisasi berbasis Light Detection and Ranging (LIDAR) masih memiliki keterbatasan dalam identifikasi lokasi objek relatif terhadap perangkat LIDAR dan pengenalan lebih lanjut terhadap jenis objek. Informasi keberadaan objek saat ini hanya berupa bounding box tanpa klasifikasi lebih lanjut, yang menghambat pergerakan manuver dan perencanaan jalur kendaraan listrik otonom. Oleh sebab itu, pada penelitian ini dilakukan pengembangan metode deteksi dan klasifikasi objek berbasis data LIDAR dengan konsep PointPillars, yang memproses data secara real-time menggunakan konvolusi 2D dimana metode ini lebih efisien dibandingkan metode yang menggunakan konvolusi 3D. Studi ini diterapkan di lingkungan terbatas sekitar jalan perkantoran BRIN Bandung, menggunakan dataset point cloud dari enam objek utama yaitu human, wall, car, cyclist, cart, dan tree. Dari total 2423 dataset, 90% digunakan untuk pelatihan dan 10% untuk pengujian. Pengembangan PointPillars melibatkan desain arsitektur backbone, komposisi layer, ukuran voxel, dan pengaturan hyperparameter. Delapan konfigurasi model PointPillars dihasilkan, menggunakan backbone 2D BaseBEVBackbone dan BaseBEVResBackbone. Konfigurasi BaseBEVBackbone dengan layer 4;6;6 dan ukuran voxel 16000 menunjukkan kinerja optimal, mencapai akurasi minimum 95%, presisi minimum 84%, recall minimum 82%, f1-score minimum 85%, dan nilai mAP 82%. Hasil ini menunjukkan bahwa metode PointPillars yang dikembangkan mampu mendeteksi dan mengklasifikasi enam objek dalam lingkungan terbatas, sehingga berkontribusi signifikan pada peningkatan fitur klasifikasi dan deteksi kendaraan listrik otonom BRIN; ---------- The development of object detection systems for autonomous electric vehicles using Light Detection and Ranging (LIDAR)-based localization methods is currently limited in accurately identifying the location of objects relative to the LIDAR device and in further recognizing object types. Existing information on object presence is confined to bounding boxes without further classification, hindering maneuvering and path planning of autonomous electric vehicles. Therefore, this study proposes a novel method for object detection and classification based on LIDAR data using the PointPillars concept, which processes data in real-time using 2D convolution, proving more efficient compared to 3D convolution methods. This study was conducted in a constrained environment around the BRIN Bandung office roads, utilizing a point cloud dataset of six main objects: human, wall, car, cyclist, cart, and tree. From a total of 2423 datasets, 90% were used for training and 10% for testing. The PointPillars development involved designing the backbone architecture, layer composition, voxel size, and hyperparameter settings. Eight PointPillars model configurations were produced, using the 2D BaseBEVBackbone and BaseBEVResBackbone backbones. The BaseBEVBackbone configuration with layers 4;6;6 and a voxel size of 16000 demonstrated optimal performance, achieving a minimum accuracy of 95%, a minimum precision of 84%, a minimum recall of 82%, a minimum F1-score of 85%, and an mAP value of 82%. These results indicate that the developed PointPillars method effectively detects and classifies six objects in a constrained environment, significantly contributing to the enhancement of classification and detection features for BRIN's autonomous electric vehicles.
Item Type: | Thesis (S1) |
---|---|
Additional Information: | https://scholar.google.com/citations?user=F41IglUAAAAJ&hl=en ID SINTA Dosen Pembimbing: DIAN ANGGRAINI: 6681986 MOCHAMAD IQBAL ARDIMANSYAH: 6658552 |
Uncontrolled Keywords: | PointPillars, Deteksi dan Klasifikasi Objek, BaseBEVBackbone, BaseBEVResBackbone, LIDAR, PointPillars, Detection and Classification Objec |
Subjects: | L Education > L Education (General) Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software T Technology > T Technology (General) |
Divisions: | UPI Kampus cibiru > S1 Rekayasa Perangkaat Lunak |
Depositing User: | Alif Ilman Nafian |
Date Deposited: | 23 Aug 2024 02:44 |
Last Modified: | 23 Aug 2024 02:44 |
URI: | http://repository.upi.edu/id/eprint/118757 |
Actions (login required)
![]() |
View Item |