1. Ekstrak Model Elevasi Digital
dalam bentuk Point Cloud dari
sensor kamera optik
Dany Laksono
January 30th, 2020
2. Extraksi Point Cloud DEM dari Kamera Optik
• Prinsip fotogrametri, SfM-MVS
• Batasan SfM-MVS
• Model Elevasi Digital dan Point Cloud dari SfM
• Beberapa Perangkat Lunak untuk SfM-MVS
4. “The art, science, and technology of obtaining reliable
information about physical objects and the environment
through processes of recording, measuring, and interpreting
photographic images and patterns of recorded radiant
electromagnetic energy and other phenomena”
Definisi Fotogrametri
--ASPRS
6. Dengan perkembangan bidang
computer vision, pengolahan citra
digital memegang peranan penting
dalam rekonstruksi 3D dari foto
untuk keperluan fotogrametri
Ulman (1979) mengusulkan metode
untuk memperoleh informasi 3D
objek tanpa informasi lain selain
gambar objek tersebut
Structure from Motion Photogrammetry
Meskipun demikian, baru pada tahun 2006, Noah Snavely
mendemonstrasikan penggunaan metode sfm untuk rekonstruksi 3D point
cloud dalam ukuran cukup besar
7. Alur Rekonstruksi 3D SfM-MVS
Structure from Motion (SfM) merupakan
metode untuk estimasi posisi dan orientasi
kamera serta gambaran kasar dari objek yang
difoto (≈ AT dan BA)
MVS (Multi-view Stereo) merupakan metode
untuk memperoleh gambaran detil objek
dari tiap piksel yang ada pada foto dengan
menggunakan hasil SfM sebagai masukannya
14. Structure from Motion (SfM) Photogrammetry
Image from Noah Snavely
The SfM Problem:
Given a set of
corresponding points in
two or more images,
compute the camera
parameters and the 3D
point coordinates
15. Structure from Motion (SfM) Photogrammetry
Image from Noah Snavely
Estimate:
- Structure xi
From:
- Camera Motion Riti
Iteratively
16. Principles of SfM Reconstruction: Epipolar Geometry
“Given point P in a multiple-view stereo images, the point P, Cl and Cr lies in
an epipolar plane where Cr ≈ Cl * f(R,t)”
17. Principles of SfM Reconstruction: Epipolar Constraint
Titik X dapat berada pada
titik manapun sepanjang
garis OL-X. Dengan
menambahkan tampilan
kedua (OR), lokasi tepat
titik X dibatasi sepanjang
garis epipolar yang
bersesuaian
Rotasi (R) dan translasi (t)
antara kedua view dapat
dinyatakan dengan
Matriks Essensial (E)atau
Matriks Fundamental (F)
18. Principles of SfM Reconstruction: Minimizing Reprojection Error
Setelah x dan R, t awal diketahui, selanjutnya dapat dihitung:
Iterasi dengan meminimalkan reprojection error:
= Bundle Adjustment
42. Sparse Reconstruction
Feature detection, matching and initial bundle adjustment resulted in
estimated camera position and initial 3D geometry of the target
Result of SfM:
Estimation of object’s
3D geometry
(or Sparse Model)
and
Camera Position and
Orientation
44. Alur Rekonstruksi 3D SfM-MVS
Jika seluruh posisi kamera
dan estimasi posisi objek
telah diketahui,
permasalahan yang
tersisa adalah
menentukan kedalaman
titik pada seluruh piksel
Algoritma Multi-view
Stereo (MVS) dapat
digunakan untuk
menyelesaikan
permasalahan tersebut
48. MVS: Depth Map to Point Cloud
Untuk tiap foto, dapat dicari depth map sebagai representasi kedalaman tiap piksel
Dari semua Depth Map di tiap foto kemudian dapat dibentuk point cloud dari objek
(structure) yang dicari
49. MVS: Depth Map to Point Cloud
Untuk tiap foto, dapat dicari depth map sebagai representasi kedalaman tiap piksel
Dari semua Depth Map di tiap foto kemudian dapat dibentuk point cloud dari objek
(structure) yang dicari
51. Dense Reconstruction: Semi Global Matching (SGM)
Semi-Global Matching (SGM) can be
based on high resolution imagery and
each pixel can be processed to render a
3-D Point Cloud of high density (150 to
380 points per square meter)
Up to 9 points per Square Meter
LiDAR
~ 172 points per Square Meter at 3” GSD
SGM
LiDAR derived point clouds are limited by the
sampling frequency of the sweeping beam and
the laser beam width (1 to 9 points per square
meter)
Wilson (2016)
52. Dense Reconstruction: Semi Global Matching (SGM)
A LiDAR derived 3-D Model is lower resolution and thus lacks sufficient resolution to
see surface changes that Semi-Global Matching (SGM) detects
However, current SGM software renders surfaces that have more low-level noise
than LiDAR surfaces
SGM DerivedLiDAR Derived
Images from the paper SEMI-GLOBAL MATCHING: AN ALTERNATIVE TO LIDAR FOR DSM GENERATION? By S. Gehrke, et al,
53. Multiview Stereo: Meshing and Texturing
“Mesh” is
obtained from
further
interpolation of
dense point cloud
Texture are obtained by ‘draping’
photo into the 3D Model based on
model’s Normal Linehttp://www.gris.tu-darmstadt.de/projects/mvs-texturing/
58. Deliverables of SfM-MVS
Sorted by the order of
processing, the result of SfM
and Multiview Stereo are
(Laksono, 2016):
a) Sparse Point-Cloud
b) Dense Point Cloud
c) Colorized Dense Point
Cloud
d) Mesh Surface
e) Textured Surface
59. DEM or Digital
Elevation
Model are
obtained from
interpolated
Dense Cloud or
3D Mesh
Produk SfM
For Mapping purpose, DEM and Orthophoto Mosaic might be more
desirable
60. DEM or Digital
Elevation
Model are
obtained from
interpolated
Dense Cloud or
3D Mesh
Other Deliverables
An Orthophoto Mosaic is an orthographic projection imagery
(“Top-Down looking” camera)
Orthophoto is
a map ready
production,
showing
elevation-
corrected
photo instead
of just Photo
Mosaic
61. Photogrammetry Produces DSM instead of DTM
General Rule:
“What the
camera (or
pixel) could
see, would be
produced as
3D”
Image from charim.net
78. Since SfM relies on Feature
Detection, it is prone to error on
feature with High Similarity
79. Limitations of SfM-MVS
Group of images
capturing eastern wing
of UGM building
Group of images
capturing western wing
of UGM building
Connectivity Graph
showing clustered
images, where it
shouldn’t be connected
at all
80. Effect of Similar Features
https://cvg.ethz.ch/research/symmetries-in-sfm/
83. Rolling Shutter Correction
Pengaturan Kamera
Apabila digunakan drone dengan tipe
copter (Mis. DJI Phantom series),
centang 'Enable rolling shutter
compensation' untuk koreksi efek
rolling shutter
rolling shutterglobal shutter
85. Perangkat Lunak SfM-MVS
Bemis et al. (2014). Ground-based and UAV-Based photogrammetry: A multi-scale, high resolution mapping tool for structural geology and
paleoseismology. Journal of Structural Geology
90. FOSS SfM-MVS
openmvg.readthedocs.org
OpenMVG
• Perangkat lunak SfM opensource berbasis Linux
• Menggunakan Algoritma Incremental SfM dan Global SfM
• Memiliki fungsi Georeference yang terintegrasi
• Tersedia GUI untuk Windows: Regard3D