28
Optimal Sensor Management Technique For An Unmanned Aerial Vehicle Tracking Multiple Mobile Ground Targets Negar Farmani, Liang Sun, Daniel Pack Unmanned System Lab The University of Texas at San Antonio

Optimal Sensor Management Technique For An Unmanned Aerial Vehicle Tracking Multiple Mobile Ground Targets

Embed Size (px)

Citation preview

Slide 1

Optimal Sensor Management Technique For An Unmanned Aerial Vehicle Tracking Multiple Mobile Ground TargetsNegar Farmani, Liang Sun, Daniel PackUnmanned System LabThe University of Texas at San Antonio

1

Outline:IntroductionProblem Statement Optimal Sensor ManagementCoordinates TransformationsGimbal Pose SelectionExperimental ResultsConclusion

IntroductionTracking multiple mobile targets in optimal manner

Previous works

Novel optimal sensor management method

Increasing number of applications of UAVs- there is lack of techniques to track multiple targets in some optimal manner when the resources (sensors) are limited to fully carry out theMissionusing a Recursive Least Square filterA* method

Camera, limitations of the resolution, the range, and the field of view (FOV)Objective of paper: present optimal technique to manage a sensor to track multiple mobile targets3

Outline:IntroductionProblem Statement Sensor ManagementExperimental ResultsConclusion

sample scenario

Specifications:assume targets move on ground randomly.UAV : constant altitude and constant velocity Control input: bank angles Camera: Limited FOV random noise. The objectives: geo-localize ground targets minimize the error of estimation on ground.

5

Outline:IntroductionProblem Statement Sensor ManagementExperimental ResultsConclusion

UAV And Target Relation

Associated Coordinate FramesThree frame of interest:Body Frame = (ib , jb , kb)Gimbal Frame= (ig , jg , kg) Camera Frame= (ic , jc , kc)

Three frame of intrest8

Target Projection

Target project into camera frame and translate to pixel locations(ex,ey)Gimbal pointing direction is determined by aligning optical axis of camera to desired direction9

Target Geo-Localization

Estimation Technique

Extended Kalman Filter (EKF)Motion model:

State of target:

Measurement model:

Geo-localizationExtended Kalman Filter (EKF): Prediction Step:

where

Geo-localization Measurement Update:

where

Candidate Generation for Gimbal Pose

Dynamic weighted graph of targets

DWG: To determine an optimal gimbal pointing direction for the purpose of minimizing the overall uncertainty of targetsDWG: represents the connection among targetsd: estimated distanceSigma: estimated position varianceDWG: computed in each iterationIth colum: estimated density of targets near target i14

Candidate Generation for Gimbal PoseCheck FOV Limited FOVCamera gimbal pointing direction

DWG generates candidates for MPCFinding min leads to optimal gimbal pose .ground location that matches the center of selected sensor FOV as destinationUAV trajectory: 15

Target EstimationsGimbal CandidatesMPC TechniqueTarget EstimationsDWGFOV TestGimbal CandidatesMPC TechniqueSharma & Pack method:Proposed method:

Sharma & Pack MethodR. Sharma and D. Pack, Cooperative Sensor Resource Management for Multi Target Geo-localization using Small Fixed-wing Unmanned Aerial Vehicles. in Proc. AIAA Guidance, Navigation, and Control (GNC) Conference, American Institute of Aeronautics and Astronautics, 2013.Develop a vision based cooperative sensor fusion technique to geo-locate multiple mobile ground targets usingDevelop a cooperative sensor resource manager using Model Predictive Control

Candidate Generation for Gimbal PoseModel Predictive Control(MPC)

Outline:IntroductionProblem Statement Sensor ManagementExperimental ResultsConclusion

Experimental Results

Experimental Results

Experimental Results

Experimental ResultsAVERAGE GEO-LOCATION ERRORS OF FIVE TARGETS FOR THE 100 EXPERIMENTS USING THE PROPOSED METHOD AND THE ONE REPORTED IN [9].Target No.12345North Position(m)15.4910.7324.8324.4615.26North Position(m)[9]11.7812.5426.3231.923.12East Position(m)13.586.0814.288.717.6East Position(m)[9]9.6910.2816.3813.711.41

Experimental ResultsOverall ErrorError (m)Improvement (%)Overall Error in North Position (m) Overall Error in North Position (m)[9]18.1521.1414

Overall Error in East Position (m) Overall Error east Position (m)[9]9.9411.9116

OVERALL AVERAGE OF GEO-LOCATION ERRORS

Experimental Results

Conclusion And Future WorksA new sensor management technique for UAVs tracking multiple targetsA dynamic weighted graphA Model Predictive Control technique

Future WorkMultiple UAVs cooperatively tracking multiple targets

Questions?

Thank You