What exactly goes on in the commercial drone/UAV industry in Singapore and globally? Behind the hype of consumer “selfie” drones lies a vast number of interesting commercial applications, where drones become an enabler for enterprises to gain new aerial perspectives of their facilities and estates, to make intelligent decisions incorporating this additional dimension of data.
In this presentation, we will look at one such drones-at-work application to reveal some of the behind-the-scene processes and technologies employed. Specifically, we will dive into the precision agriculture domain and share some of the computer vision problems we face, and take a look at various potential solutions to these challenges.
Call Girls Jalahalli Just Call 👗 7737669865 👗 Top Class Call Girl Service Ban...
Garuda Robotics x DataScience SG Meetup (Sep 2015)
1. Drones At Work ::
Capturing Data, Generating Insights,
Solving Real Problems
Ong Jiin Joo
CTO, Garuda Robotics
jiinjoo@garuda.io
DSSG - 23/9/2015
2. Drones at work gathering useful data
(1) Not
flying
for
fun
(2) Not
flying
and
shoo5ng
for
aesthe5cs
Drones
at
work:
(1) Solve
Customer’s
Problems
(2) Gather
Useful
Data:
Can
be
analyzed
to
produce
intelligence
/
insights
3. In the next 45 minutes
• Share our experience
– Behind the scenes
– Technology and processes
• Data capture workflow
• Data analysis workflow
• Precision agriculture case
study: Tree counting
4. Case Study Background
Running
example
in
this
presenta1on:
Precision
Agriculture
for
Palm
Oil
Planta5ons
Planta1on
customers
want
to
know:
How
many
trees
are
there
in
my
planta5on?
This
affects:
(i)
Manpower
&
equipment
planning,
(ii)
fer1lizer
purchase
and
dissemina1on
15. Urgency of analysis
• When do we need the deliverable
– Real Time or within minutes / hours
– Non Real Time (days / weeks)
• Some analysis require huge amount of
compute – such as image recognition
16. Tradeoff between using more bandwidth to
transport data elsewhere vs. shipping more
compute power on site
In-‐country
Telco
Ground
Sta1on
Drone
Cloud
Services
Wi-‐Fi
3G
Dongle
Internet
backbone
18. Data Storage Size
Photogrammetry Example
(simplified)
• Fly at 100m, Camera FOV 90° both
sides, 1 picture covers 200x200m = 4 ha
• Suppose plantation 10,000 ha square (or
10km by 10km)
• 80% overlap required ~= shooting 5
times same area
• Total size: (10,000/4) * 5MB * 5 = 62.5GB
– fits one 64GB SD card.
20. Data Presentation :: Image Stitching
• Combine
everything
or
by
blocks
• Highly
repe11ve
• Lack
control
points
21. Data Presentation :: Orthomosaics
• Geometrically
corrected
• Can
be
placed
on
map
Used
by
surveyors
to
measure
true
distance
22. Data Presentation :: 3D Reconstruction
• Photogrammetry
methods
Similarly,
used
by
surveyors
to
measure
length,
area
and
volume
of
interest
in
3D
space
25. Descriptive Analytics
Example:
Telco
Tower
Inspec5on
• Is
the
antenna
s1ll
slanted
at
2.8
degrees
from
ver1cal?
• Any
disconnected
wires,
bird
nest,
damage
from
harsh
weather?
Example:
Flare
Stack
Inspec5on
• Is
the
structural
integrity
of
the
flare
stack
holding
up?
• Is
the
flare
stack
opera1ng
at
normal
temperature?
26. Predictive Analytics
What will happen next?
Example: Solar Panel
• What is wrong?
• How many times
observed
• Correlate with
electricity yield curve
27. Back to our case study ::
Plantation Management
Dry
leaves,
but
next
to
river.
Why?
Empty
space,
but
no
palm
planted.
Why?
Winding
road,
difficult
to
bring
harvested
palm
out.
Redo?
Palm
of
mixed
age:
high
maintenance
cost.
Is
it
1me
to
replant
the
en1re
area?
If
so,
should
the
river
be
shi]ed
for
water
to
drain
be^er?
28. Case Study :: Plantation Management
Great!
Now
I
just
have
to
keep
it
going
for
25
years
How
much
fer1lizers
do
I
need
to
get?
How
should
I
distribute
them
so
that
my
workers
don’t
just
throw
excess
away?
How
many
trees
do
I
have!?
30. Tree Counting
The image cannot be displayed. Your computer may not have enough memory to open the image, or the image may have been corrupted. Restart your computer, and then open the file again. If the red x
still appears, you may have to delete the image and then insert it again.
• More advanced
ways
1. Satellite imagery
2. Drone imagery
3. Apply Computer vision
31. Tree Counting
Posi1ve
Features
Histogram
of
• Colour
• Intensity
• Mean
• Standard
Devia1on
Nega1ve
Features
(random
sampling)
Our
naïve
model!
RFC
33. Ways to improve tree counting
• Non-CV techniques
– Operations: capture trees at same size and
light intensity (vary altitude, time of flight etc.)
– Domain info: planting patterns, tree distance,
max tree per block
– Past data: information from previous flights,
manual count, last count
• How about CV techniques?
34. Ways to improve tree counting
Source:
Oil
Palm
Tree
Detec2on
with
High
Resolu2on
Mul2-‐Spectral
Satellite
Imagery
h?p://www.mdpi.com/2072-‐4292/6/10/9749?trendmd-‐shared=0
13
April
2014
35. Ways to improve tree counting
Active research area
• Some new proposals
• Undergoing R&D and
trials with our corpus
• Trials with customer
with existing data
about their tree count
36. Tree Counting :: Next Steps
• Impact from good tree count
– Yield prediction and correction
– Plantation ops
– Prescriptive Analytics together with Arborists
• Next things to classify
– Healthy trees vs. sick trees
– Other trees / crops
– Heterogeneous plantations
37. Summary
• Drones are already at work
delivering actionable insights
• We can capture the data with
our drones, but the challenge is
to go beyond the descriptive
into the predictive and
prescriptive analytics
• Lots of opportunities coming
soon