Error in Tilt Angles (Actual vs. Predicted)

-90

-85

-80

-75

-70

-65

-60

-55

-50

-45

-40

-35

-30

-25

-20

-15

-10

-5

0

110 115 120 125 130 135 140 145 150 155 160 165 170 175 180

Time (s)

Tilt

Ang

le

TILT TILT PREDICT 1 TILT PREDICT 5

Error in Normalized Pan Angles (Actual vs. Predicted)

0

10

20

30

40

50

60

70

80

90

100

40 45 50 55 60 65 70 75 80 85 90

Time (s)

Nor

mal

ized

Pan

Ang

le

PAN PAN PREDICT 1 PAN PREDICT 5

ABSTRACTABSTRACT

PREDICTIVE TRACKING ALGORITHMPREDICTIVE TRACKING ALGORITHM

Visual Visual Tracking of an Tracking of an

Unmanned Unmanned Aerial Vehicle Aerial Vehicle (UAV) Using (UAV) Using

GPSGPSSamuel S. StarrSamuel S. Starr

Emir TumenEmir Tumen

Advisor: Dr. George PappasAdvisor: Dr. George PappasSpecial Thanks: Ben Grocholsky, Jim KellerSpecial Thanks: Ben Grocholsky, Jim Keller

SOFTWARE FLOWCHARTSOFTWARE FLOWCHART

The project provides a tracking system for The project provides a tracking system for Unmanned Aerial Vehicles (UAVs), which are Unmanned Aerial Vehicles (UAVs), which are used for research and security purposes and used for research and security purposes and require real time control. Giving the camera the require real time control. Giving the camera the ability to follow an object, and limiting the ability to follow an object, and limiting the margin of error in the visual tracking output, the margin of error in the visual tracking output, the project aims to satisfy the need of an automated project aims to satisfy the need of an automated surveillance for UAVs at the University of surveillance for UAVs at the University of Pennsylvania. In the chosen approach, the GPS Pennsylvania. In the chosen approach, the GPS information from the UAV is collected and information from the UAV is collected and stored in the ground base station PC. The GPS stored in the ground base station PC. The GPS data is then fed to the Pan-Tilt unit’s control PC, data is then fed to the Pan-Tilt unit’s control PC, which processes this information and converts which processes this information and converts it to pan and tilt values by using a geometrically it to pan and tilt values by using a geometrically based algorithm. Finally, two servomotors based algorithm. Finally, two servomotors which accept these pan and tilt values adjust to which accept these pan and tilt values adjust to point the camera in the UAV’s direction. The point the camera in the UAV’s direction. The current implementation of the system is well current implementation of the system is well suited to any visual output device such as suited to any visual output device such as monitors, televisions or webcams for network monitors, televisions or webcams for network sharing, and for real-time visual tracking of any sharing, and for real-time visual tracking of any GPS transmitting object in a 360° field.GPS transmitting object in a 360° field.

Group #6

UAV GPS DataFeed

GPSlocation of

CameraBase

Synchonizedat

BeginningRS-435 or 802.11 connection

Output to Robotic PT Mechanism via SerialCable (RS-232)

Microsoft Visual C++Coding of Tracking

Algorithm(Simple or Predictive)

PTUTracker.DLL(Unmanaged DLL File)

PTUWrapper.DLL(Interface with ROCI module)

ROCI Interface (Access to all GPS/Position feeds GRASP Lab has)

Conversion to Universal Transverse Mercator (UTM) Performed

ROCI Module

UAV GPS DataFeed

GPSlocation of

CameraBase

Synchonizedat

BeginningRS-435 or 802.11 connection

Output to Robotic PT Mechanism via SerialCable (RS-232)

Microsoft Visual C++Coding of Tracking

Algorithm(Simple or Predictive)

PTUTracker.DLL(Unmanaged DLL File)

PTUWrapper.DLL(Interface with ROCI module)

ROCI Interface (Access to all GPS/Position feeds GRASP Lab has)

Conversion to Universal Transverse Mercator (UTM) Performed

ROCI Module

Demo Times:Demo Times: 1030h, 1100h, 1130h, 1300h1030h, 1100h, 1130h, 1300h

PENN UAV HARDWARE OVERVIEWPENN UAV HARDWARE OVERVIEW

CloudCap Piccolo Onboard AvionicsCloudCap Piccolo Onboard Avionics (10 Hz GPS, 100Hz IMU)(10 Hz GPS, 100Hz IMU)

¼ Scale Piper J3 Cub UAVs¼ Scale Piper J3 Cub UAVs

Position DifferencesCalculated

dLat, dLong, dAlt

Scales into Pan and Tilt Commands for RoboticPT Mechanism

Multiplies Degree calculations by 19.44445

Calculation of PanAngle

atan(sqrt(CLat2 +CLong2)/CAlt)

Calculation of TiltAngle

atan(CLat/CLong)

Prediction AlgorithmUsing Position and Velocity Data to

Predict Future Location of UAV

UAV Position & VelocityData Feed

Prediction Offset Calculated

Vlat/Freq, Vlong/Freq,Valt/Freq

Calculate Control/Predicted Position VectorCLat = dLat + Vlat/(k*Freq), CLong = dLong + Vlong/

(k*Freq), CAlt = dAlt + Valt/(k*Freq)

Look-Ahead Time = LAT = 1/(k*Freq of Data)

Assuming No System Delay: k=1 LAT = 1/FreqFor Our System: LAT = Approximate Net System Delay

(1 second)

SYSTEM FLOWCHARTSYSTEM FLOWCHART

• Provide a live visual video output of an object that provides three-dimensional GPS position and velocity data.• Operate in a look-ahead fashion in which the system will anticipate where the vehicle is located in the one-second future.• Interface the camera output with any type of visual display (both IN and OUT)• Provide output visible throughout the entire operation of the UAV. (max. 30 minutes)• Tracking algorithm is compatible with the maximum velocity of a UAV. (20 meters/second)

SYSTEM SPECIFICATIONSSYSTEM SPECIFICATIONS

The Predictive Tracking Algorithm was originally derived using vector theory. While this approach could have worked, it proved unnecessarily complex. An alternative was a geocentric approach, which proved to be too sensitive to any changes in latitude and longitude. The simple geometric approach adopted eliminates substantial sensitivity

errors, but is limited in that it works best in close range and only once GPS is

converted to UTM. It is the conversion to UTM that allows any geometric algorithm to work precisely. The Look Ahead Time can be adjusted to predict the position of the

moving object within the delay time.

Error in Normalized Pan Angles (Actual vs. Predicted)

-100

-90

-80

-70

-60

-50

-40

-30

-20

-10

0

10

20

30

40

50

60

70

80

90

100

0 15 30 45 60 75 90 105 120 135 150 165 180 195 210 225 240 255 270 285 300

Time (s)

No

rma

lize

d P

an

An

gle

PAN PAN PREDICT 1 PAN PREDICT 5

Error in Tilt Angles (Actual vs. Predicted)

-90

-85

-80

-75

-70

-65

-60

-55

-50

-45

-40

-35

-30

-25

-20

-15

-10

-5

0

0 15 30 45 60 75 90 105 120 135 150 165 180 195 210 225 240 255 270 285 300

Time (s)

Tilt

An

gle

TILT TILT PREDICT 1 TILT PREDICT 5

Sources of Predictive Error

• Increase in UAV Velocity• Increase in System Delay

• Quality of GPS Data• Mechanical System

Limitations

SIMULATION TESTING OF ALGORITHM ERRORSIMULATION TESTING OF ALGORITHM ERROR

UAV DATA

GEOMETRICALGORITHM

[latd, longd, altd],[latv, longv, altv]

[Pd, Td]

VideoOutput

[Pf, Tf][latp, longp, altp]

Up to10 Hz

Look Ahead Time [LAT] (sec.)

Error% 0.2 1 5

PAN 0.8900 4.3860 35.9149

TILT 0.2828 1.4425 7.6173

Based on the above graphs, and the Based on the above graphs, and the adjacent chart, the algorithm that adjacent chart, the algorithm that

minimizes error for system delay and is minimizes error for system delay and is less sensitive to changes in UAV velocity is less sensitive to changes in UAV velocity is when LAT is set equal to the delay on the when LAT is set equal to the delay on the

order of 1 second.order of 1 second.