The 2016 Remote Sensing Field camp will take the form of two projects.
A low tech, low cost aerial photography project using visible spectrum UAV/Ultralight Aircraft mounted cameras as the sensor to demonstrate that relatively low tech, low cost solutions can achieve surprisingly good results when compared to more commercial systems.
A more high tech, high cost terrestrial LiDAR collect of a building or structure of historical or architectural significance.
The scope of a project will influence all other aspects of the project, including its cost, timing, quality and risk.
1. Remote Sensing Field Camp 2016:
Comparing Photogrammetric and
LiDAR 3D Modeling Techniques
By: Emma McLeod, Maria Swindells, Ryan Poirier, Doug
Tymchuk, Margie Massier & Stuart MacDonald
2. Project Outline:
Purpose & Objectives
Flight line Planning
Positional Control
Laying Targets
Real Time Kinematic (RTK)
Total Station
Aerial Imagery Acquisition
Building 3D Models from Oblique Aerial Photos
Comparing Aerial Derived 3D Models to the Inspire UAS (quadcopter)
models
Comparing Aerial Mosaic to Ebee Data Collect
Terrestrial LiDAR Acquisition and 3D Models
Difficulties embedding 3D Models on the web
Comparison of Projects
3. Purpose:
The purpose of the remote sensing field camp was to
compare both 3D photogrammetric and LiDAR models of
COGS derived from point clouds to determine if a 3D
photogrammetric model is comparable to LiDAR modeling
methods.
4. Objectives:
Objective 1 – Mosaics: Aerial data collect of COGS property
including Lawrencetown exhibition grounds with both Ultra-light
and Ebee Survey Drone. Creating two seamless ortho mosaics to as
high a positional standard as can be achieved and validated against
RTK GNSS derived check points for comparison.
Objective 2 – 3D Models: Aerial photography collect and UAV data
collect of the exterior of the COGS building and the Lawrencetown
exhibition grounds for 3D modelling in a web browser. Both model’s
accuracy will be compared with LiDAR derived control.
Objective 3 - LiDAR: A closed traverse laser scan of the exterior of
the COGS building including the road surface and curbs that can be
imported to Faro Scene and validated against RTK GNSS / total
station derived positional control points.
Objective 4 – Presentation: Present the results of the field camp
at the AGEOS student presentations in the COGS AV room on
Thursday the 26th of May. A forty-five minute presentation to be
delivered detailing the processes, results and recommendations
derived from the projects.
6. Flight Planning
Overview:
Purpose:
Allow photogrammetry students to become more familiar with flight plan
management and to determine what, if any non proprietary flight planning
software can be utilized and possibly implemented into the
photogrammetry course in the future.
Planning:
Create flight plan that will achieve the Remote Sensing Field Camp’s goal’s
of creating a seamless, color balanced mosaic for Low/High Res and
Low/High cost comparison to the Ebee survey drone mosaic.
7. Flight Planning will use - TopoFlight
Software Suite and Addition Spread
Sheet Data to Compute Flight Lines,
Image Footprints and Positional Control
Trial License Provided by Klaus Budmiger at Topoflight.com for the Remote Sensing Field
Camp 2016
8. Identify Client Needs/ Sensor and
Platform Capabilities
Client – Remote Sensing Department/ COGS
Needs – Achieve the highest possible resolution and Stereo
coverage with platform and sensor provided for the 2016 remote
sensing field camp
Sensor and Capability - Ricoh GX200. Image capture rate is 1
image per 5 seconds. Focal length of 5.1 to 33.5. Sensor image
size in pixels 4000 x 3000. Pixel Size 1.86 microns – 0.00867mm.
Back-Up Camera - GoPro 4. Camera: 41.0mm height, 59.0mm
width, 29.6mm depth Camera with housing: 71.3mm height,
71.1mm width, 39.0mm depth
Platform Capability - Fixed wing Ultra-Light. Average Flying
speed is 45knots/80 kph
9.
10. Import Spreadsheet Parameters for
Flight Line Creation
Pixel Size, Altitude , Minimum and Maximum Forward/Sidelap, Minimum Sensor Capture Rate, and Average
Flying Spead.
26. Leica Total Station
A Leica brand total station was used to determine the coordinates of some
of the corners of the COGS building, in order to validate accuracy of the
imagery and 3D data collected. The total station can be set up over a
known or unknown point, and then a prism is held in the corner of the
building so the laser can shoot from the total station to the prism in order
to record the distance and angle. This enables a more accurate reading
than using a GNSS where the location would be scattered due to the
proximity of reflective, built up surfaces.
27. The total station is first set up over a known or unknown point, on a level
tripod. Known locations (survey monuments) are then shot to calculate
the distance and angle from the total station so the total station can
determine its location and direction. From there, points can be surveyed
using the survey prism, and the points and distances are recorded as an
ASCII file which can be transferred off the instrument when survey is
complete.
Leica Total Station
30. Agisoft PhotoScan 3D Model
Processing Oblique Aerial Photos to produce a 3D Model of COGS
and the Lawrencetown Exhibition grounds
31. Loading and Aligning Photos
High Accuracy was chosen to produce the
highest quality result. This parameter takes a
lot of time but is necessary because as
Accuracy decreases it deteriorates the quality
by a factor of 4.
Pair preselection was disabled. It speeds up the
process to have it enabled, but chooses pairs
based on a lower accuracy setting and may
affect results.
The point limits were left as the default
parameters.
33. Geo-referencing and Accuracy Assessment of
Photogrammetric derived 3D Models
Setting the reference coordinate
system to match the coordinate
system of gathered RTK points to be
used as GCP markers.
Imported a CSV file containing UTM
Easting and Northing coordinates for
RTK locations on the COGS building.
Added and refined marker placement
on each image used to generate a
georeferenced point cloud. This
improved the accuracy of the
produced orthomoasic.
34. Building a Dense Cloud
Medium Quality was chosen when
building the dense cloud as it takes the
average of nearby points producing a
smooth result. On higher qualities the
product becomes messy looking.
Depth of filtering was set to Aggressive
because the product does not contain
meaningful small details. If this
parameter is set to Mild then it will
produces a result with edges that
appear wavy.
Building a dense cloud in PhotoScan calculates a point cloud
similar in density to a LiDAR point cloud. It can then be edited,
classified and exported from PhotoScan much like LiDAR.
36. Building a Mesh
Building a Mesh connects the point cloud creating a solid 3D image.
Arbitrary Surface was chosen to highlight
the buildings within the imagery.
Source Data is set to Dense Cloud.
Face count is automatically calculated
based off the number of points in the dense
cloud, but can be modified.
Interpolation was enabled to avoid manual
hole filling at the post processing step.
38. Building Texture
In Generic Mapping mode no assumptions
are made regarding the type of scene
and creates the most uniform texture
possible.
Mosaic Blending mode is used in all
multiphoto projects.
The default was left for Texture size and
count. In circumstances when there is
limited RAM lower resolutions to many
files can be substituted to acquire a high
resolution the same level of resolution.
49. Terrestrial LiDAR Scanning
Terrestrial LiDAR uses a near
infrared laser to scan a subject
(in this case the COGS building)
to create a point cloud/3D
model.
Terrestrial LiDAR is known for its
high precision and accuracy.
50. Components of a Terrestrial LiDAR
Scanner
Scanner
Mirror
SD Memory
Card Slot
Start / Stop
Button
Display
Screen
Power
Button
Example settings used
for the COGS rooftop
51. LiDAR Scan Process Overview:
1. A scan plan was created for potential scan locations around the
school and on the roof
2. Parameters were researched which would provide the most
accurate results.
3. The scan of the school was completed in 19 scans
4. A new FARO SCENE project was created and scans were imported
into the project
5. Scans were manually placed using the correspondence view in the
software
6. Scans were registered together using the place scans “cloud to
cloud” method.
7. Scans were georeferenced using the total station data.
52. LiDAR Scan Plan & Parameters Used
Scanner Settings Used for the COGS Scan
Outdoor …20m Used outdoors when the distance
between the scanner and the object of
interest is less than 20 meters.
Resolution Setting 1/5 – based on the level of detail
needed, settings used for the scan are
for outdoor/large spaces
Quality Setting 4x – Quality settings are based off
environmental conditions, a lower
quality setting is required when
conditions are good.
Scan with Colour ON
Sensors ON
53. Manual Scan Placement:
Correspondence view in FARO
SCENE shows the rough
registration locations for each of
the scans used.
Manual scan placement entailed
placing the scans around their
approximate location to other
scans.
Placement had to be precise, if
scans were not placed properly
then the scans would not register
together properly.
54. Scan Registration:
Once the scans were placed around
their approximate locations, cloud
to cloud registration was used to
tighten up the scan placement.
Cloud to cloud registration is
commonly used for the refinement
of already positioned scans,
different initial positions can lead to
different results.
55. Georeferencing the Scans:
Scans were referenced into UTM
coordinates by inputting the UTM
position from the GPS into the
previous lat/long coordinates
under the transformation tab.
Total station data was imported
into the Faro SCENE software as a
.csv file
Scan 008 was used as a reference
scan as it contained the majority
of the total station points
Sample total station data with
approximate Easting, Northing and
Elevation values
56. Georeferencing:
2 markers were placed in scan 008
(in planer view) where known total
station points were located.
To determine the transformation
needed the coordinates of both the
known total station points and the
markers were retrieved and the
differences between them were
calculated.
The differences were applied to
the total transformation for all
scan values which resulted in
proper georeferenced scans.
58. Shortcomings faced during the scan
Automatic target based registration did not work – the
spheres were unable to be recognized by the SCENE
software so manual registration had to be used.
Weather was also an issue faced during the project, as
the LiDAR Faro scanner required a clear day without
moisture on the ground.
59. Embedding 3D data in a Website
Using VRML to be embedded in HTML with Cortona 3D plugin
- Problem 1: New Google Chrome Browser does not support Cortona
3D plugin
- Problem 2: The VRML file produced by Terrestrial LiDAR is >35 Gb
and is not supported when brought into Firefox, Microsoft Edge, or
Explorer
X3D (successor to VRML)
- FaroScene does not export to this file type
- Must be converted from VRML, however file size makes the
conversion impossible using conversion apps
FaroScene WebShare
Provides clients with a web address to view a point cloud, however this
method is expensive.
CloudShare was investigated as well however proved expensive at up to
$166/month
61. Aerial and eBee Mosaic Comparison
Aerial Ebee UAV
Inexpensive ($10,000 for plane, or
~$200 an hour)
Expensive ($30,000 UAV, or ~ $160 an
hour)
Requires technician to process through
PhotoScan
Comes with automated software
Requires pilot and insurance coverage Requires a certified UAV commercial
piolet license
GoPro used has low resolution at 1500
ft (issues with Ricoh camera with
better resolution during this flight)
Very high resolution due to high
quality sensor and low flight height
64. Model Comparison
Photogrammetric Models LiDAR Models
Inexpensive (Go Pro and charter flight
<$1000)
Expensive ($75,000 sensor and highly trained
technician to acquire and process data)
Low accuracy High accuracy measured to millimeters
Best results occur on a cloudy day without
rain or snow
Best results occur on a sunny or cloudy day
without rain
Georeferencing is done manually and takes
some time
Easily georeferenced with internal GPS system
Processing can take anywhere from 25 min
to 30 hrs
Depending on the scale of the project, multiple
scans can be time consuming
65. Cloud Compare Software Results
Unfortunately, the Cloud
Compare software was
unable to compare the
two different point clouds
due to the size of the
LiDAR project as well as
the inability to match the
scale of the two projects.
66. Final Thoughts on 3D Models
The type of modelling method used for 3D analysis is
dependent on the needs, resources and budget of the user.
Hiring new people in all projects and will likely be necessary
and should be taken into consideration (LiDAR Tech, UAV tech
or pilot)
Time of day and weather are important for each type of
collection which makes planning difficult. Having a flexible
crew and buffer time is imperative when completing these
types of projects.
If a user requires a high accuracy and detailed model where
measurements will be taken from the model, then a LiDAR 3D
solution may be more suited.
If the user requires a low cost model where a high level of
detail is not necessarily needed then a 3D photogrammetric
model may be more appropriate and cost effective.
Inspire UAV was also used to collect and gave an intermediate
result between the two
67. Special Thanks to:
Jim Norton
Rob Hodder
Paul Illsley
Bernie Rector of East Coast Ultra Light
John Saunders of John Saunders Photography & Film
Editor's Notes
Emma
Maria
Emma
Margie, ryan stew doug
ryan
Maria
Stew
Margie
Margie
Emma
Emma
Maria
Maria
Maria
Doug
Maria
Stuart
Emma
Emma
Emma
Known as a transverse scan plan – this is because adjacent scans are registered using targets placed in the overlapping areas.