This document provides an overview of UAV image recognition technology and applications. It defines UAVs and describes the key technologies that have enabled their development, such as autopilots, GPS, and miniaturized components. It outlines the UT UAV group's work on autonomous target recognition for competition, including detecting, analyzing, and determining the position of targets in images. The group's system achieves 85% detection accuracy and aims to reduce position error below 50 feet. Potential applications of UAVs discussed include monitoring oil pipelines and ranches as well as aiding wildfire response. Strict regulations govern UAV use due to safety concerns.

1. UAV Image Recognition
Technology and Applications
The UT UAV Group
Cockrell School of Engineering
The University of Texas at Austin
Harmony Mones‐Murphy
Chockalingam Viswanathan
Tejas Kulkarni

2. What is a UAV?
• The Department of Defense Dictionary
defines a UAV as:
A powered, aerial vehicle that does not carry a
human operator, uses aerodynamic forces to
provide vehicle lift, can fly autonomously or be
piloted remotely, can be expendable or
recoverable, and can carry a lethal or nonlethal
payload.

3. Driving Technology
• Powered heavier than air flight
• Radio control (R/C)
• Autopilots
• GPS
• Imagery systems
• High density power batteries
• Long range and low‐power micro radio devices
• Miniaturized parts
• Wireless networks
• Powerful micro‐processors

4. 1898 1912 1933 1960
1903 1918 1959
HISTORICAL FIRSTS

5. 1898: First demonstration
of radio‐control
Nikola Tesla’s “Teleautomaton,” a radio‐
control boat
Electrical Exposition at Madison
Square Garden

6. 1912: First autopilot
Elmer and Lawrence Sperry
Curtiss B‐2
• A gyrostabilizer hydraulically
operated the elevators and rudder.
• Allowed the aircraft to fly straight
and level without pilot input.

7. 1918: First radio‐controlled unmanned flight
• Forerunner of the
modern cruise missile.
Curtiss‐Sperry Aerial Torpedo

8. 1959: First unmanned reconnaissance aircraft
Northrop Radioplane SD‐1 Falconer/Observer

9. TYPES OF UAVS

10. Fixed‐Wing
Northrop Grumman RQ‐4 Global Hawk

11. Rotorcraft
Helicopter: Northrop Grumman MQ‐8 Fire
Scout
Quadcopter Tiltrotor: Bell Eagle Eye

12. PAYLOADS

13. Electro‐optic Payload Systems
• Optical Cameras
• Low‐light‐level
(LLL) Cameras
• Thermal Imagers

14. Radar Imaging Payloads
• Synthetic
Aperture
Radar(SAR)

15. Dispensable Payloads
• Military – Missiles
• Civil ‐ Pesticides

16. UT UAV

17. Our Team
• Undergraduate
• Interdisciplinary
• Student leadership

18. AUVSI Competition
• Student UAS
Competition in
Maryland
• Reconnaissance
mission
• Fourth year of
participation
• 1st in Autonomous
Target Recognition in
2010

19. Phoenix II
Airframe
Avionics
Imagery

20. UT UAV Overview
• Our Implementation
– Target Detection
– Target Analysis
– Position Determination

21. Target Characteristics
• Position (LLA)
• Background Shape
• Background Color
• Alphanumeric 4 to 8 feet
Character
• Alphanumeric Color
• Orientation
4 to 8 feet

22. Target Detection
• Color‐based approach
– Outlier image
• Exploit target attributes
– Size, aspect ratio
• Implemented on DSP
– Texas Instruments
C6748

23. Background Image
• Represent image in 3‐D color space
– ,
• Image contains background and
foreground

24. Foreground Image
• Average RGB pixels in frame
0.06
Red Plane
0.04
P e rc en t
0.02
0
0 50 100 150 200 255
Green Plane
0.04
P erc e nt
0.02
0
0 50 100 150 200 255
Blue Plane
0.04
P erc e nt
0.02
0
0 50 100 150 200 255
Pixel Intensty
• Compute distance from mean
• Distance threshold determines potential
targets

25. Outlier Image
• Potential targets highlighted in oultlier
image
Original Image Outlier Image

26. Binary Image
• Remove noise ‐ windowed median filter
• Label objects ‐ connected component
Binary Image Label Image

27. Target Analysis
Bounding
Segmentation Skeleton Compare
Rectangle Rotate
Cropped
Image

28. Target Detection Performance
• Tested on scaled airfield and recorded
video
– Robust to trees, runways
– Poor at detecting some colors
Specification Performance
Speed 10 frames per second
Detection Accuracy* 85%
False Positive Rate 10%
* Accuracy = ratio of targets detected to total number of targets

29. Results
Legend: Correct Incorrect Marginally Incorrect

30. Target Position Determination
• Convert image coordinates to absolute position
• Position Accuracy
– Maximum allowable error – 150 feet
– Desired error – less than 50 feet
• Monte Carlo Error Analysis
– Sweep camera 60 degrees in all directions from the
vertical
– Estimate standard deviation of error

31. Monte Carlo Error Analysis
Error Analysis (500 feet altitude) Standard Deviation (feet)
1000
750 45
500 40
250
35
feet
0
30
-250
25
-500
20
-750
-1000 15
-750 -500 -250 0 250 500 750 1000
feet

32. System Overview
Sony FCB
EX‐980S
Target Analysis
LabVIEW
Texas
Instruments
C6748
Triangle J Purple
Yellow NW Lat Lon

33. Plans for 2012
• Communication
– Switch to Wifi (802.11N)
• Digital camera (DSLR)
• Weight reduction

34. Why UAVs?
• UAVs are suited for doing the “dull, dirty
and dangerous” tasks of everyday life.

35. Applications of UAVs in Texas
• Oil & gas
• Wildfires
• Ranching

36. Oil & Gas
Use to check pipelines for leaks
Use as a method to collect and transmit
data between rigs

37. Wildfires
Bastrop County Wildfire
Aid Firefighters with real time
information and firefighting
capability.

38. Ranching
Spraying crops with pesticide and
fertilizer, monitoring crops, soil,
moisture, and pest conditions, and
insect sampling
Use to track cattle/deer
Check fences for holes

39. Safety
• Due to safety concerns there are strict
regulations regarding the use of UAV’s in
unrestricted airspace throughout the
world.

40. Air Systems Lab
• All the work done in the Air systems lab
are undergraduate student projects, for
various competitions.

41. Q&A