Armazeta, Aura Jovita Gandari (2026) Analisis Perbandingan Ortofoto 3D Gaussian Splatting Dan SfM/MVS Untuk Rapid Assessment Kerusakan Bangunan Berbasis OBIA Pada Bencana Banjir Bandang (Studi Kasus: Kabupaten Pidie Jaya, Aceh). Masters thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 6016251026-Master_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
6016251026-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (20MB) | Request a copy |
Abstract
Rapid assessment pasca bencana banjir bandang membutuhkan informasi spasial yang cepat dan akurat untuk mendukung pengambilan keputusan. Pemetaan UAV memungkinkan akuisisi data resolusi tinggi dalam waktu singkat, namun pengolahan ortofoto dengan metode Structure from Motion/Multi-View Stereo(SfM/MVS) membutuhkan waktu komputasi yang lama. 3D Gaussian Splatting (3DGS) muncul sebagai alternatif yang berpotensi menghasilkan rekonstruksi lebih cepat. Data foto udara diolah menjadi ortofoto menggunakan dua metode, SfM/MVS dan 3DGS, yang dibandingkan berdasarkan efisiensi komputasi, uji ketelitian, kualitas radiometrik dan visual, serta stabilitas permukaan air. Ortofoto 3DGS selanjutnya digunakan untuk ekstraksi informasi kerusakan bangunan melalui OBIA, meliputi segmentasi, klasifikasi tutupan lahan, klasifikasi kerusakan bangunan berbasis rule-based, dan uji akurasi. Hasil penelitian menunjukkan metode 3DGS unggul dari segi efisiensi waktu (54 menit–1 jam 37 menit, dibanding SfM/MVS 5–7 jam) dan stabilitas permukaan air, dengan uji ketelitian (RMSE horizontal 0,299–0,368 m) dan kualitas radiometrik-visual yang sebanding dengan SfM/MVS. Klasifikasi tutupan lahan dengan Random Forest menghasilkan akurasi tertinggi (OA=92,8%; Kappa= 0,908), termasuk pada kelas Bangunan (93,18%/96,24%). Klasifikasi kerusakan bangunan berbasis skor kumulatif menghasilkan akurasi sangat baik (OA=92,31%; Kappa=0,883), dengan kategori Rusak Berat cenderung underestimation akibat keterbatasan sudut pandang nadir. Hasil ini menunjukkan integrasi UAV, 3DGS, dan OBIA mampu mendukung rapid assessment kerusakan bangunan pascabencana secara cepat dan akurat.
=====================================================================================================================================
Post-disaster rapid assessment of flash floods requires fast and accurate spatial information to support decision-making. UAV-based mapping enables high resolution data acquisition within a short time, yet orthophoto processing using the Structure from Motion/Multi-View Stereo (SfM/MVS) method requires considerably long computation time. 3D Gaussian Splatting (3DGS) has emerged as an alternative method with the potential to produce faster reconstruction. Aerial photographs were processed into orthophotos using two methods, SfM/MVS and 3DGS, which were then compared based on computational efficiency, accuracy test, radiometric and visual quality, and water surface stability. The 3DGS-based orthophoto was subsequently used for building damage information extraction through OBIA, comprising segmentation, land cover classification, rule-based building damage classification, and accuracy assessment. The results show that the 3DGS method outperforms SfM/MVS in terms of processing time efficiency (54 minutes–1 hour 37 minutes, compared to 5–7 hours for SfM/MVS) and water surface stability, with accuracy test (horizontal RMSE of 0.299 0.368 m) and radiometric-visual quality comparable to SfM/MVS. Land cover classification using Random Forest yielded the highest accuracy (OA = 92.8%; Kappa = 0.908), including for the Building class (93.18%/96.24%). Building damage classification based on a cumulative scoring approach achieved a very good accuracy (OA = 92.31%; Kappa = 0.883), with the Heavily Damaged category tending toward underestimation due to the limitations of the nadir viewing angle. These results indicate that the integration of UAV technology, 3DGS, and OBIA is capable of supporting rapid assessment of post-disaster building damage both quickly and accurately.
Actions (login required)
 |
View Item |