Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

The YellowScan Surveyor:

5cm Accuracy Demonstrated

Pierre Chaponnière1 and Tristan Allouis2


1
Application Engineer, YellowScan
2
CTO, YellowScan

Additional 2 years of R&D development and


Introduction intensive component testing led to the conception of
YellowScan Surveyor, the very latest lightweight the YellowScan Surveyor. This latest product
UAV borne lidar system developed by YellowScan concentrates the best high-end components with the
was tested in real time conditions in the field. The highest accuracy/weight ratio available on the
aim of this paper is to provide detailed information market.
on the level of accuracy that can be expected from
the system and the methodology adopted to assess YellowScan Surveyor
it. The YellowScan Surveyor includes:
The YellowScan Surveyor lidar system was flown on  The Applanix APX-15 UAV, a single board
a multicopter UAV platform over an area surveyed GNSS-inertial solution chosen for its
by a professional certified surveyor. Detailed unprecedented post-processed position and
comparisons between the two acquisition methods orientation accuracy in its category
are described in this paper.  The Velodyne VLP16 (also known as Puck), a
dual return laser scanner
YellowScan’s turnkey solutions  An onboard computer for continuous data
acquisition and processing
YellowScan, a French company has developed
 Battery (up to 1.5 hours autonomy)
lightweight professional grade lidar systems made
 Worldwide technical and operational support
for UAVs since 2012. Prototyping phases evolved
into a viable commercial solution, the Mapper – a
fully integrated self-powered 2.1kg lidar system – YellowScan Surveyor key specifications:
which started to be commercialized in 2014.  Laser scanner frequency: 300 kHz
 Weight: 1.6 kg, battery included
The 4 key principles behind each YellowScan system  Power consumption: 15W
are:  Autonomy: 1.5 hours typical
 Turnkey solution
 Size (mm): 100x 150 x 140
 Ultra-light weight
 Independent from the aerial platform
 Simple to use

The YellowScan Surveyor: 5cm Accuracy Demonstrated 1/6


from a national private network of automated
Study Site and Dataset GNSS base stations (TERIA). 5 readings were
taken per control points achieving a RMS3D of
Study site description
5mm.
The site selected for demonstrating the accuracy
 From this 5 points baseline grid, a professional
level obtained with the YellowScan Surveyor is
certified surveyor installed a Leica total station
YellowScan’s calibration site located in Assas, south
TCRP 1201+ and shot multiple points over the
of France, directly north of Montpellier.
surveyed field including table corners and table
It consists of a 150x60m model aircraft training site top flat surfaces, airstrip gravelly surfaces and
which exhibits gravelly airstrip and a 40m long row natural grassy ground (see Figure 1). Polygonal
of 9 concrete tables (1x2.3m) together with relatively error to setup on the baseline grid was +/-10mm
sparse low and high vegetation. The overall terrain is and points were subsequently picked up at
mainly flat with occasional drainage ditches (20 to +/- 2.5mm accuracy.
50cm deep) surrounding the air field.
Overall 70 points were picked up and used for the
lidar point cloud validation. This included:
GCP - Survey methods to establish
control points  36 materialized points suitable for checking
XYZ accuracy. Those points correspond to table
In order to establish the most accurate validation
corner points (4 corners for each of the 9 tables)
points for comparison purpose to the lidar dataset
the following method was conducted:  34 ground points for checking Z accuracy, in
details these can be classified as:
 5 ground points were materialized at the edges
of the surveyed ground using wooden beacons ➢ 9 flat table surface points (picked up in the
and surveyed using a double frequency middle of the table)
GPS/GLONASS Septentrio APS NR2 receiver
➢ 14 ground points picked up along the
receiving RTK corrections via GSM connection
gravelly airstrips of the site

Figure 1. Ground surveying of the validation points

The YellowScan Surveyor: 5cm Accuracy Demonstrated 2/6


➢ 11 ground points located on natural grassy
Method
surfaces in the vicinity of the tables.

System set-up Flight plan parameters


The flight plan adopted for this survey included 8
The YellowScan Surveyor system was installed on
flight lines, 3 flown centered and along the table
the OnyxStar Fox-C8 multirotor UAV (Figure 2)
row, 3 flown perpendicular to the table row and the
which has a flight time of about 25min and a take-
last two flown as a cross-like figure with one axis
off weight of 9kg.
along the table row and the other perpendicular to
Lever arm measured and entered into the Applanix it. The flight line spacing was set at about 20 to 25m,
system via Ethernet cable communication. The fly height at 20m and speed at 3m/s (see Figure 3).
Surveyor was switched on 5min prior to the flight to
Overall flight time recorded for the mission was
optimize satellite acquisition and locking procedure.
5min and the total area covered summed up to
Prior to the flight the YellowScan team set up a about 0.5ha.
Septentrio GNSS base station APS-NR2 next to the
The flight also displays two bathtub geometries
surveyed field on a fixed control point (see GCP
flown at 8m/s at the start and at the end of the flight
description above). This was left reading 30min
which are recommended as the initialization
prior to the flight and stopped at the end of the
procedure for aligning the heading measurement of
mission. The logged data was saved and used
the Applanix GNSS-Inertial unit.
subsequently during the PPK process of the
trajectory.

Figure 2. YellowScan Surveyor onboard the OnyxStar Fox-C8

The YellowScan Surveyor: 5cm Accuracy Demonstrated 3/6


Lidar Point cloud generation Automated shape detection method for
As per standard post processing procedure when XYZ accuracy estimation
operating the YellowScan Surveyor, the trajectory In order to exclude human / subjective intervention
was imported into POSPac together with the raw in materializing the table corners within the lidar
data (RINEX file) gathered by the Septentrio base point cloud used to check XYZ accuracy, an
station. automated approach was favored.
POSPac operates corrections to the trajectory and Using Terrasolid software (TerraScan,
enables the user to specify – whenever possible – the TerraModeler), lidar points corresponding to the
exact coordinates the base station was setup at. In tables were classified as a specific ‘table’ class and
our case we specify its coordinates picked by the used as an input class within the automated
surveyor together with the offset of the tripod. ‘vectorize building’ function (Roof class ‘table’,
The corrected trajectory together with boresight maximum gap of 1m, planarity tolerance of 0.05m,
calibration angles of the Surveyor lidar unit were increase tolerance of 0.05m minimum area of 1m²,
finally used to generate the corresponding point minimum detail of 0.5m² and max roof slope of
cloud (las format) using YellowScan’s processing 5deg). Corners of the produced shape were
software. digitized, their XYZ coordinates extracted and
compared to the validation points.
The Velodyne Puck laser scanner records 360deg
field of view. For the purpose of this document, a Accuracy estimation method
30deg corridor along each flight line was kept in the
To assess the difference between the lidar point
point cloud while the rest of the points were
cloud and the validation points we used the Root
classified as overlap and thus discarded for the
Mean Square Error – a well-known and vastly used
accuracy estimation.

Figure 3. Flight plan trajectory

The YellowScan Surveyor: 5cm Accuracy Demonstrated 4/6


accuracy estimator. RMSE (1) takes the difference
between the observed and the estimated/forecast
Results
value for each user-defined point, squares it, finds
Point cloud parameters
the mean squared value and generates the square
root of that mean. The generated point cloud (Figure 4) for the 0.5ha
surveyed area totalizes 21M points for a disk space


N
1 of 580Mb. These parameters led to a point density of
RMSE= N ⋅∑ ( f i −oi )2 (1)
i=1
about 6000points/m² with all flight lines combined
Where : and translated into a measured density of
N =number of forecast / observation pair 1600points/m² per flight line.
f = forecast
o=observation
Z validation
Two types of validations were produced from this The results obtained from the Z-validation process
survey: are summarized in Table 1.

 A Z-only validation which looks at the vertical Considering all the points, the calculated RMSE is
distance between the control point and a 2.1cm. The average error, also called absolute
triangulated surface generated directly from the accuracy or bias, is -1.3cm. The standard deviation,
point cloud (Terrasolid parameters during the also called precision or repeatability, is 1.6cm.
output control point report being ‘maximum The type of material seems to affect the precision
triangle length to 20m, maximum slope to 20deg level: as notionally expected, the smoother the
and Z tolerance to 0.2m). surface the higher the precision. The elevation of flat
 A XYZ validation which looked at comparing cement slabs is identified and mapped with an
the 4 corners of each of the 9 tables using an RMSE of 1.9cm, an accuracy of -1.6cm and a
automated shape detection method described precision of 1cm. Gravelly airstrip and natural
before. grassy terrain measurements present an accuracy of

Figure 4. Resulting point cloud

The YellowScan Surveyor: 5cm Accuracy Demonstrated 5/6


-1.3cm and -1.2cm respectively, and precision of and introduce errors in the generation of the table
1.6cm and 2.2cm respectively. shape during the Terrasolid step.

XYZ validation
Conclusion
Table 2 summarizes the results (in m) of the
The YellowScan Surveyor used in parallel with an
comparison between the 36 table corner points
onsite GNSS base station reaches absolute accuracy
picked by ground total station and the 36 table
of -1.3cm in Z with a precision of 1.6cm. As
corner points automatically generated in Terrasolid
expected, the harder the surface is, the higher the
software.
precsion will be. In the study, cement slab surfaces
The average error (absolute accuracy) for X, Y and Z are positioned with an accuracy of -1.6cm and a
is 2.5cm, -3.3cm and -2.6cm respectively. The precision of 1cm.
standard deviation (precision or repeatability) for X,
Assessing XYZ accuracy is relatively more
Y and Z is 3.3cm, 3.8cm and 0.9cm respectively. The
demanding in terms of target material and
RMSE for X, Y and Z is 4.1cm, 5cm and 2.8cm
automatic shape generation technique in order to
respectively.
exclude subjective interaction. However, the study
The materialized corner edges of the tables have a shows an average error below 3.3cm and a precision
higher XY error than a Z error. This is mostly due to below 3.8cm.
the fact that edges tend to be noisier than surfaces

Table 1. Z validation

Table 2. XYZ validation

The YellowScan Surveyor: 5cm Accuracy Demonstrated 6/6

You might also like