Volume 5, Issue 3, March – 2020 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Orthomosaic from Generating 3D Models with

Photogrammetry
Abbas Z Khalaf1 , and Akram Hameed2
1
Assist. Prof Dr. Department of Civil Engineering, Geomatic Branch, University of Technology, Baghdad, Iraq.
2
Msc, student, Department of Civil Engineering, Geomatic Branch, University of Technology, Baghdad, Iraq.

Abstract:- This paper, especially discuss the getting of II. THE WORKFLOW TAB IN AGISOFT
three coordinates for points in two images by processing
these images to 3D models by Photogrammetric Agisoft do the arrangement of processing by Workflow
program. Agisoft is openly utilized functions that tab. The processing is done by navigating during the tab and
developed algorithm used in a field called Computer executing all tasks below;
Vision which needed for processing images to 3D models.
Computer Vision is a sub-field of computer science that  Adding Images
was prepared to help robotic equipments to clarify - In this stage, images are loading to the program.
surrounding bodies. Computer Vision algorithms detect  Aligning Images
common points among photos of a given objective, - In this stage, the searching of finding the same pixel
possessed from various angles, and process those photos between images and guess the position of camera and object
to conclude depths of the structures. There are four basic model from matching.
steps: Align Photos, Build Dense Cloud, Build Mesh, and  Dense Cloud Building
Build Texture and two secondary steps: Build DEM, - After completing the second phase, just a portion of the
Build Orthomosaic. Additionally, there are a number of points is handled to a dense point in 3D space.
tasks with related sub-settings of the basic procedures of  Mesh Building
Agisoft that developed by author to refine the results of - In this stage, links every group of three neighboring points
these processes. Regrettably, user guide of Agisoft has into a trigonal face, that join easily to product a mesh above
not been used correctly in the factors and the effect of the exterior of the body.
each sub-setting. In order to quite understand the  Texture Building
concepts and functions of sub-settings to refine the - This stage interested in built-up the texture depending on
output results, the author reached an experiment in main first images and cover the mesh to reach for the real
which he created Photogrammetric model of the building object
were processed using various sub-settings. It is not as  DEM Building
simple as some program, but it have more better cog of - This is the penultimate step in which heights are positioned
all step of operation, and produce much more accurate for the model based on images.
results.  Orthomosaic Building
- This is the last step, in which an image of the model is
I. INTRODUCTION taken and this image contains the coordinates for each
location in the image.
Photogrammetry is a measurement method where the So as to get better outcome, there is a different way into
coordinates of the points in 3D of an object are computed the program, except the basic procedures, this paper take the
from the measurements made in two photographic images main and sub-setting workflow.
(or more) taken starting from a different position.
III. CAPTURING AND ADD IMAGES
Generally, the Photogrammetry can be divided into
two types the aerial photogrammetry and terrestrial The first step is obtaining the data,take images with
photogrammetry. The aerial photogrammetry is used for your camera and then place them on your computer to use in
developing particular and public grounds and production the software.
of topography maps. On the other hand, the terrestrial
photogrammetry can be used for surveying on the ground All imaging work was done with one camera Canon
level for small scale, manufacturing and industrial EOS 500D Furthermore the image dimensions was
applications. Terrestrial is also defined as digital close range 4752x3168 pixels. Moreover the acquired images are in the
photogrammetry (DCRP). Recently, the using of digital form of pairs that g100% overlap and hence more efficient
photogrammetry becomes considerably better comparing coverage of the object. With manual setting, the focal
with traditional photogrammetry, since all images are now a length is 50 mm over that, the (image stabiliser) mechanisms
digital format, furthermore, due to the low cost of digital was averted in work of imaging to preserve stability, where
cameras, storage media and processing systems. The digital all The listed pixel positions must illustrate a repeatable
image is used in different fields and applications. measure for point and the shutter speed was f/5.

IJISRT20MAR0109 www.ijisrt.com 48
Volume 5, Issue 3, March – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
So the ISO increased if more sensitivity is necessary in ISO was 100 with no flash.
low lighting case, for the condition of this study the used

Fig 1:- Left image and Right image

IV. ALIGNING IMAGES The second step of Align images sub-setting is Key
Point Limit. This setting allocates how many pixels Agisoft
It is the step to give the images for the software to uses from every image to detect features in the photo. The
compare the images and find the matching points between default setting of Key Point Limit is 40,000 (for reference, a
the images by an auto matching process illustrate in figure typical 36 Megapixel image has approximately 240,000
(2), which the software is based on functions that let the user pixels, and a 21 Megapixel image has approximately
to get coordinates ( x , y ) in pixel of matching points in 180,000 pixels). Selecting a high number for Key Point
each image , illustrate in table (1) . This is the fascinating Limit will rise the working time until the value reaches
component of the Photogrammetry process, without that 240,000, providing full processing of images having 36
nothing else would run. Megapixels (240,000 pixels), (stated in One Agisoft forum
discussion). However, the forum also illustrated that if the
After uploading images into the software, the first step setting selected for Key Point Limit is more than 40,000,
of a constructed model is Aligning images. The software there will be no refinement in the quality of Dense Clouds in
allows to user to choose the accuracy of matching from low next processing. so ,if accuracy is the main regard,
to highest passing through classes between them. Highest projection error can be reduced to 0.3 pixels by rising Key
accuracy uses full resolution image, and Lowest accuracy Point Limit to 120,000 (although a rate projection error with
uses 10% resolution. The next sub-setting is Pair Pre Key Point Limit set to 40,000 is 0.7 pixels--already less than
Selection; this sub-setting allows to Agisoft do a quick pre- 1 pixel) (Marcel, 2015). The author chose the Align images
scan before it starts full common points detection in the process with Key Point Limit: 40,000 (default) and Key
image. This quick pre-scan can find two images that share Points Limit: 90,000 (Fig. 3 and 4). There is no noticeable
the same views. Depend on this pre-scan, Agisoft can skip difference in the development of the formed sparse points.
the entire image point detections to align each single photo. The outcome must be a scattered points for the body.
Instead of, it only scanning two pre-selected images every
time to find camera position. As a result, use Pair Pre
Selection will reduce processing time.

Fig 2:- Auto Matching between images

IJISRT20MAR0109 www.ijisrt.com 49
Volume 5, Issue 3, March – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig 3:- Align images process with Key Point Limit: 40,000 (default) setting. Total number of sparse points were 3,645

Fig 4:- Align images process with Key Point Limit: 90,000 (default) setting. Total number of sparse points were 3,665

As we note there is no noticeable difference in the points and in the second case there is 3665 spare points,
number of spare points , where in the first case 3645 spare With a significant and tangible variance in Key Point Limit.

Left image Right image

ID X Y ID X Y
1 10.985 0.579 1 9.324 0.828
2 7.041 0.383 2 4.707 0.603
3 1.874 0.23 3 -1.117 0.429
4 1.074 0.192 4 -1.898 0.395
5 8.902 0.196 5 7.195 0.439
6 8.983 0.197 6 7.283 0.439
7 4.229 0.158 7 1.316 0.365
8 4.379 0.157 8 1.476 0.363
.
.
.
.
.
.
3636 1.901 6.372 3636 -0.983 6.362
3637 2.826 6.375 3637 -0.043 6.394
3638 9.489 6.369 3638 8.39 6.69
3639 2.215 6.168 3639 -0.67 6.171
3640 4.695 6.032 3640 1.958 6.116
3641 5.039 5.994 3641 2.352 6.09
3642 6.448 5.915 3642 4.064 6.068
3643 4.701 5.875 3643 1.956 5.962
3644 5.273 5.745 3644 2.621 5.853
Table 1: Matching point between image from aligning phase

IJISRT20MAR0109 www.ijisrt.com 50
Volume 5, Issue 3, March – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
The final sub-setting of the Align images step is Tie (default) and Tie Point Limit 4,000; increasing the number
Point Limit, which allows how many detected common of tie point limits to twice more than doubled the number of
points are truly applied to calculate the depth of the scene. formed sparse points, but at a cost of a significantly longer
For instance, when 40,000 Key Point Limit and 2,000 Tie processing time (Fig. 5 and 6).
Point Limit are applied, 40,000 points of each photo are
used to detect common points to match photos, and then the Depending on the results, the Tie Point Limit value
2,000 highest quality common points for each photo are immediately impacts the overall number of sparse points.
used to estimate camera locations and to produce point The author decided not to do experiments on the Quality
clouds. Reducing the Tie points setting that lead to reduce sub-setting at this phase of processing because successful
the processing time of image alignment. In the Align images image alignment generally based on photo-shooting, not
stage the author applied a setting of Tie Point Limits: 2,000 Quality sub-settings

Fig 5:- Align Photos process with Tie Point Limits: 2000 (default) setting. Total number of created sparse points were 18,489.

Fig 6:- Align Photos process with Tie Point Limits: 4000 (default) setting. Total number of created sparse points were 18,489.

IJISRT20MAR0109 www.ijisrt.com 51
Volume 5, Issue 3, March – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
V. EDITING THE SCATTERED CLOUDS choose points and selected them in another color and
remove these points. When you reach to satisfy, then
The moment that leads to unconvinced with the structure the dense from excerpt all the useful data from
aligning of all images, it should be remove of any obviously every image.
invalid points. Use the indented-circle-shaped lasso tool to

Fig 7:- Scattered Clouds

VI. DENSE CLOUD BUILDING five settings: Ultra High, High, Medium, Low, and Lowest.
Ultra High quality uses a full size image resolution, High
In this stage, it uses the images that aligned to create a uses 50% of the original scale, Medium uses 25%, Low uses
new cloud points, which intense appropriate to build a 12.5%, and Lowest uses 6.75%. These percentages
model that appear from a distance. This stage is divided into immediately impact the total number of points in the dense
two parts: the first is to get the points resulting from the cloud (Fig. 8).
process of forward intersection to points of auto matching in
the alignment phase and their number (3644 points) as The result of this stage when quality is Ultra high was
shown in table (1), and the second part is the process of (7,164,195 points), High (2,156,384 points), Medium
dense these points, it will become their number (7,164,195 (573,862 points), Low (142,389 points), Lowest (34,884
points points ) As shown in the table (2) .It can adjust the points).
Quality rate of dense as wished for, This quality process has

Fig 8:- Dense Cloud

IJISRT20MAR0109 www.ijisrt.com 52
Volume 5, Issue 3, March – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Points Dense Cloud

ID x y z
1 -1.92114 0.146128 -3.33072
2 -1.91918 0.15541 -3.33179
3 -1.91525 0.164334 -3.32858
4 -1.92061 0.166833 -3.33162
5 -1.91757 0.126316 -3.3243
6 -1.91793 0.131492 -3.32662
7 -1.91525 0.152375 -3.3293
8 -1.92061 0.161121 -3.33197
9 -1.91846 0.162549 -3.33126
.
.
.
2138786 -1.91543 0.132028 -3.32412
2138787 -1.91936 0.142916 -3.32715
2138788 -1.9165 0.142916 -3.32733
2138789 -1.91989 0.166119 -3.33144
2138790 -1.91614 0.153089 -3.33019
2138791 -1.91846 0.12703 -3.32519
2138792 -1.91846 0.164691 -3.33179
2138793 -1.91525 0.162192 -3.32876
2138794 -1.92025 0.143094 -3.32894
Table 2:- Point Dense Cloud

The second sub-setting is Depth Filtering. This filter may distinguish blue sky as a blue object, and thus generates
has four list options: Disabled, Mild, Moderate, and point clouds in what would otherwise be void space. So ,
Aggressive. The Aggressive option means that if the object Mild works well on a monotone surface, and Aggressive
is monotone, Agisoft will make it as void space and will not works well with photos that contain actual void spaces.
generate points within that space (Fig. 9). Moderate filtering works between the previous two filtering
options (Fig. 11). The disabled mode does not distinguish
Choosing the Mild option induce the program to depth; that's mean, if an object has a monotone colored
distinguish the monotone areas as objects, and to generate surface, Agisoft does not regard it as a surface and generates
points within those spaces (Fig. 10). For instance, the void spaces (Fig. 12). In summary, all depth filtering
Aggressive depth filter does not distinguish the surface of a selections have advantages and disadvantages, thus the
whiteboard as an object, and it generates a hole in its Depth Filtering setting must be selected depending on
photogrammetric model; on the other side, the Mild setting photography conditions.

Fig 9:- Build Dense Cloud process with Depth filtering: Aggressive setting.

IJISRT20MAR0109 www.ijisrt.com 53
Volume 5, Issue 3, March – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig 10:- Build Dense Cloud process with Depth filtering: Mild setting.

Fig 11:- Build Dense Cloud process with Depth filtering: Moderate setting.

Fig 12:- Build Dense Cloud process with Depth filtering: Disabled setting.

IJISRT20MAR0109 www.ijisrt.com 54
Volume 5, Issue 3, March – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
VII. BUILDING MESH In this process the sub-settings are divided into general
options contains ( Surface Type, Source Data, Face Count),
The dense cloud appears big solid object, but seems and advance options contain ( Interpolation, and Point
actual cloud points as the name propose when zoom in it. So Classes ) ( Fig. 12). Surface Type has options Height Field
, the next step is Build Mesh (mesh is the geometry of the (2.5D) and Arbitrary (3D); when the object is a flat
generated model depending on the point cloud). The structure, Height Field (2.5D) is appropriate; but, if an
mesh will connect the point object has 3D structure, Arbitrary (3D) must be chosen (Fig.
cloud into faces to make a continuous surface. 13 and 14). Either Sparse Cloud or Dense Cloud may be
chosen from Source data. Generally, Dense Cloud is the best
since it generates more close details.

Fig 13:- Mesh sub-settings

Fig 14:- Build Mesh process with Surface Type: Arbitrary setting.

Fig 15:- Build Mesh process with Surface Type: Height Field setting.

IJISRT20MAR0109 www.ijisrt.com 55
Volume 5, Issue 3, March – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
The Face Count (Face Count means the total number of points, Medium generates 1/15 of the point cloud (1/3 of the
the generated surface of the mesh) sub-setting has High, High setting), and Low generates 1/45 of the point cloud
Medium, and Low options; High generates faces, or surface (1/9 of the High setting) (Fig. 15, 16 and 3-17).
of meshes, that are made of 1/5 of points in the Dense cloud

Fig 16:- Build Mesh process with Surface Count: High setting. The total face count is 1,438,665.

Fig 17:- Build Mesh process with Surface Count: Medium setting. The total face count is 479,554.

Fig 18:- Build Mesh process with Surface Count: Medium setting. The total face count is 159,850.

IJISRT20MAR0109 www.ijisrt.com 56
Volume 5, Issue 3, March – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Another sub-setting is Interpolation. This sub-setting
has three options: Disabled, Enabled (default), and Extrapolated generates meshes between discrete points
Extrapolated. Disabled accurately rebuilds meshes and speedily fills holes, included large holes (Fig. 20). So, if
depending on point clouds (Fig. 18). Alternatively, Enabled the aim of a Photogrammetry project is to obtain accurate,
measures a certain diameter from each point and generates the Disabled Interpolation setting is the best. On the other
meshes as if the discrete points were found within the side, when visualization is the purpose of the
diameter; in one word, it automatically fulls small holes(Fig. Photogrammetric modeling, Enabled is the preferable.
19).

Fig 19:- Build Mesh process with Interpolation: Disabled setting.

Fig 20:- Build Mesh process by Interpolation: Enabled setting.

Fig 21:- Build Mesh process with Interpolation: Extrapolated setting.

IJISRT20MAR0109 www.ijisrt.com 57
Volume 5, Issue 3, March – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
The Point Classes are the last sub-setting. This sub- mode generates a UV map for ball-like, spherical bodies. A
setting is for GIS-based mapping program; it permits to texture is a 2D plane; so, Texturing a spherical form needs a
import point cloud from mapping software. They can be unique mapping projection. However, this mode is only
assorted into various categories, such as Building, appropriate for spherical structures. Single image generates
Vegetation, Ground , Water, etc. This sub-setting permits texture depending on one image; an image can be selected
Agisoft to read the categories assigned by GIS mapping from any image in the active Chunk. Keep UV is for
software, and then generate meshes of selected point types. importing a UV map generated in various software. Many
modeling programs permit to create the texture on meshes.
VIII. BUILDING TEXTURE Blending mode is another Sub-setting. To illustrate,
Mapping Mode is applied to form UV maps, and Blending
The next step of the workflow is Build Texture. In this mode is applied to form chose photo-images on generating
process, Agisoft generates a texture and places it on the UV maps as textures. Blending mode has five options:
surface of the meshes. The first sub-setting is Mapping Mosaic, Average, Max Intensity, Min Intensity, and
Mode; this sets methods of UV mapping of generating Disabled.
textures. UV mapping generates a 2D atlas, or a canvas on
which textures can be projected. Options for the Mapping The Mosaic option is the default blending mode. This
Mode sub-setting are Generic, Adaptive Orthophoto, setting chooses the closest image to corresponding surfaces
Orthophoto, Spherical, Single Camera, and Keep UV. The and uses that image without blending with other overlapping
default setting is Generic; in this option, Agisoft images. The Average sub-setting computes an average pixel
automatically selects the best photos (depending on cubic value, or the RGB value of colors, of overlapping images
projection) and forms a UV of meshes into texture atlases so and uses those pixel values as textures. The Max Intensity
as to transform the 3D structure of meshes into a 2D image setting applies images that have maximum intensity, or
format for generating textures. This mode is preferable for bright pixels, to corresponding surfaces. The Min Intensity
more sophisticated models that have 3D structures . setting is an inverse of the Max Intensity setting; it uses
Orthophoto generates a photomosaic from a projection plane minimum intensity images. The Disable setting is, still, for
and projects a generates photomosaic onto the meshes. imported models that usually have appropriate textures. In
closing, depending on the author’s experiences, Mosaic
In another way which used by Orthophoto, the provides the best result between those sub-settings; so,
Adaptive Orthophoto discovers vertical faces and generates Mosaic is preferable for generating textures on most
textures independently for those side meshes. Spherical Photogrammetric models.

Fig 22:- Build Texture process with Blending Mode: Mosaic setting.

IX. BUILDING DEM of a modified geographical system. The coordinates system

already selected for the model itself is by default takes The
It's an important step for generating an Orthomosaic, coordinates system. It can be changed to another one, as
which allows Agisoft to produce a Digital Elevation Model. long as datum parameters can be obtained. Planar
These values appear as a surface of a neat grid of elevation. projection-type, Agifoft permits to the user put DEM on a
DEM plays an important case in surface changes. Sparse plane. Projection plane and orientation can be chosen during
point can be used to extract DEM, a dense points or a mesh. create DEM. By a group of markers, it's possible to specify
Depending on data of dense cloud more outcomes are the plane. For example, to create DEM for project interested
computed. There are four parameters that effects on Build with facades perhaps use Planar projection-type. Cylindrical
DEM, the first is projection-type this parameter divided into projection-type, permits to build DEM on a surface of
three sub-setting, Geographic - permits to select a cylindrical shape. In this state the value of height is
geographic system with available options or load parameters

IJISRT20MAR0109 www.ijisrt.com 58
Volume 5, Issue 3, March – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
computed as the measure between the model to the cylindric order to dense cloud points are rebuilt. By Enabled (default)
surface. mode: Agisoft will compute elevations for all regions that
appear in photos. The Agisoft software creates a solid
The second parameter is Source data, It is computed sample with several elevation data in Extrapolated mode.
DEM depending on dense cloud data. Elevation data results
can be created from a sparse point, averting Dense Building The fourth effect is Point Classes permits to choose a class
step to avoid losing time. or classes for generating DEM. An actual resolution of
ground for DEM evaluated by the source data displays by
The third parameter is Interpolation, If it is Disabled the resolution value.
it's getting an accurate outcome because of the regions in

Fig 23:- Build DEM.

X. BUILDING ORTHOMOSAIC The second groups of parameters that affect on

Orthomosaic is detailed below:
It is the last step after building the model in which the
coordinates ( x, y, z ) can be obtained anywhere in the  Projection type
model. An orthomosaic, also known as an orthophoto, is a  Geographic permits to select a geographic system with
photo of a region, collected of multiple photos stitched available options or load parameters of a modified
together using Photogrammetry, which has been corrected geographical system. The coordinates system already
and scaled for accuracy. Generally, orthomosaics are of selected for the model itself is by default takes The
much better resolution and detail than satellite imagery of coordinates system. It can be changed to another one, as
the same region, that reason for the use of high-resolution long as datum parameters can be obtained.
cameras at a comparatively closer distance from the ground.  Planar: Agifoft permits to the user put DEM on a plane.
Projection plane and orientation can be chosen during
There are two groups of parameters that affect on create DEM. By a group of markers, it's possible to
building orthomosaic, the first group is color calibration, the specify the plane. For example, to create DEM for
first one of color calibration is Source data that deal with project interested with facades perhaps use Planar
which type of data that consider a basis for estimation of projection-type.
overlapping regions. It is favored to choose the type of  Cylindrical: permits to project orthomosaic on a
source data depending on the surface that the orthomosaic cylindrical surface. It uses to avert severe distortions
on. This parameter divided into Sparse cloud: the when projecting cylindrical objects, like round towers,
assessment of the rough. Model illustrates more accurate tunnels, tubes etc.
outcome, but only in the case of sufficient details of the
surface. DEM is a substitution of value of Model for the  Surface
great data group when it is impossible to rebuild a mesh. Orthomosaic generation depending on the data of the
The second color calibration parameter is Calibrate white DEM is specially efficient for time saving on mesh creation
balance a further choice to be turned on if white balance step during aerial survey data processing scenarios.
was wanted. Alternatively, mesh surface type permits to generate
orthomosaic for less common applications,like creation of

IJISRT20MAR0109 www.ijisrt.com 59
 Volume 5, Issue 3, March – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
orthomosaic for models and frontispiece of the buildings The ground sampling resolution refers to the default
that might not be indicated at all. value for pixel size. The value of pixel size can be used in
meter or in other units.
 Blending mode
There are three types of blending the first is Mosaic that  Max. Dimension (pix)
determine the frequency fields where the highest frequency The greater dimension of raster data can be set by
item founded along seam line. The second is Average that this parameter.
use the mean value of pixels. The third is Disabled that
interested in the color of the pixel in images to rebuild the  Back face culling
surface. Projection of regions can be ignored on the surface
where directed the counter of normals to the targeted
 Pixel size projection plane. Where surface data are available, the
Agisoft creates orthomosaic for the total area.

Fig 24:- Build Orthomosaic

XI. NEXT STEPS AND ADDITIONAL OPTIONS XII. CONCLUSION

There are more additional procedures that can take to This paper submits two main reasons to reach precise
improve the model. results: the first is the accuracy of the camera, where all its
accuracy was better, the results were more accurate, and the
Partial Save second reason is the test of sub-setting lists that affect on the
Building 3D models take more time may be reach 4-6 number of points of auto-matching where the more points
hours when hundreds images are used in every process, the more dense will be better and therefore the mesh is
sometimes cannot complete the work simultaneously so better and the end result is the best.
Agisoft permits to save the work partially.
REFERENCES
Model scaling
The scale adds manually, When that can see the same [1].Marcel, 2015, Alignment Experiment.
points at least from three images, this helps to add a scale, [2].Agisoft LLC, 2014, Agisoft PhotoScan User Manual:
by putting points called markers then compute the distance Professional Edition, Version 1.4.
between the points. [3].Kotaro Yamafune, 2016, USING COMPUTER VISION
PHOTOGRAMMETRY (AGISOFT PHOTOSCAN)
Exporting the Model TO RECORD AND ANALYZE UNDERWATER
For exporting any model after finishing it, there are SHIPWRECK SITES.
different type of formats for display or importing such as [4].Agisoft Online Forum, 2019.
Autodesk, VRML: This format allows you to work with the
MATLAB environment, and more than 10 formats to use.

IJISRT20MAR0109 www.ijisrt.com 60