Working towards scenario-based evaluations of first responder positioning systems

Transcription

1 Working towards scenario-based evaluations of first responder positioning systems Jouni Rantakokko, Peter Händel, Joakim Rydell, Erika Emilsson Swedish Defence Research Agency, FOI Royal Institute of Technology, KTH Precision Indoor Personnel Location and Tracking Technology Workshop Worcester, 1-Aug-2011

2 Robust multisensor positioning/navigation GPS Accelerometers Gyro Magnetometers Barometric altimeter Individual sensors - GPS - IMU (foot- or backmounted) - Magnetometer - Baro-altimeter - Other possibilities include Doppler-radar, cameras, laser, ultrasound,

3 GPS-denied environments Position error based on individual sensor set will inevitably increase with time/movement Ranging between first responders and cooperation reduces error drift but we still need regular recalibration of positions Possibilities Radio-based ranging to anchor nodes Building floor-plans Satellite imagery and ranging to recognizable objects, image data base matching,

4 Background

5 Why Test and Evaluate (T&E)? Experimentally based research helps identifying and formulating relevant research challenges, and helps ensuring end-user needs are targeted adequately T&E enables the continued development of - positioning algorithms and sub-systems - methodology (through training) - presentation technologies, man-system-interaction aspects Evaluations of performance for existing systems Requirements definition, challenging yet realistic req s Procurement support, incl. acceptance testing

6 Why Test and Evaluate (T&E)? Experimentally based research helps identifying and formulating relevant research challenges, and helps ensuring end-user needs are targeted adequately T&E enables the continued development of - positioning algorithms and sub-systems - methodology (through training) - presentation technologies, man-system-interaction aspects Evaluations of performance for existing systems Requirements definition, challenging yet realistic req s Procurement support, incl. acceptance testing

7 On the need for T&E Examples of existing PDR-type positioning systems Step detection, step length estimation, heading estimation System Horizontal Vertical Weight [g] Size [cm] Power [W] X < 2% of dist. < 1.5m < 0.5 Y < 2% of dist. < 1.5m N/A Z < 2% of dist. < 1.5m

8 On the need for T&E Examples of existing PDR-type positioning systems Step detection, step length estimation, heading estimation System Horizontal Vertical Weight [g] Size [cm] Power [W] X < 2% of dist. < 1.5m < 0.5 Y < 2% of dist. < 1.5m N/A Z < 2% of dist. < 1.5m Is this the truth, the whole truth and nothing but the truth?

9 On the need for T&E How should Safety-of-Life (SoL) critical systems be tested? How should the performance of SoL critical systems be presented? - refined figure of merits required, that are more complex and realistic, including dynamic scenarios, varying environments, - simple example the car fuel consumption metric which is based on different driving cycles? Accurate reference systems needed for evaluations

10 Scenario-based evaluations The way the first responder moves affects positioning accuracy Walking, running, crawling, surface type, Accuracy depends upon scenario Benefits with radio-based ranging and cooperation varies with scenario, e.g. ranges, amount of multipath, attenuating objects (walls), nr of units, GPS accuracy deteriorates in urban and indoor environments Scenario-based evaluations with actual first responders are required But they must be complemented with other tests

11 Testing facilities & tools Facilities KTH R1 Urban Combat Infantry Training School Evaluation systems VICON-lab at Linköping University Ubisense UWB RTLS Camera-based reference system

12 Camera-based reference system Goal Provide reliable ground truth data using a minimum of preinstalled infrastructure during scenario-based T&E Background General camera-based positioning is a difficult problem, requiring tracking of a large number of landmarks and/or recognition of previously visited areas as the camera moves though a scene By installing visual markers at known positions, an easier problem is obtained

13 System description Preinstall visual markers in area of interest Place camera e.g. on helmet or shoulder Localization using ARToolKit++ is performed in two steps estimate global position (which marker) estimate local position (location relative marker)

14 Evaluation Decimeter accuracy typically Error increases Close to center of marker Distances over a few meters x - axis y - axis z - axis Red VICON system Blue ARToolKitPlus Difference

15 Evaluation

16 Summary Results Accuracy < 2-3 decimeters Range < 2-3 meters The reference system provides adequate performance for use in typical indoor situations (e.g. corridors, door openings), and at transitions indoor-outdoor Future improvements Low-visibility, longer range, near-ir

17 KTH R1

18 KTH R1 Sweden s first experimental nuclear reactor site

19 KTH R1 Instrumentation Accurate indoor positioning reference system being installed - Pre-installed radio-based (UWB) positioning system - Base-stations estimates positions of tags - Combined AoA and TDOA positioning

20 KTH R1 Possibilities Algorithm and sub-system development and evaluation Scenario-based testing of existing and experimental indoor positioning systems

21 Urban Combat Training Site A new (small) city center at the Urban Combat Training Center outside Linköping, up to company level exercises Shops, restaurants, hotel, family homes and apartments ~ 1km 2, 44 houses, 111 apartments, fully furnitured Possibilities Camera-based reference system for scenario-based evaluations Urban op s, outdoor-to-indoor transitions, and pure indoor op s

22 VICON lab at Linköping University VICON motion capture system Utilizes near infrared cameras and strobes along with reflective markers Multiple cameras covers a room, approx. 6-by-8 meters in size Expensive but extremely accurate (sub-mm) Usability Algorithm and sub-system development and evaluation Foot-mounted IMU with markers (DLR, Germany)

23 What needs to be tested? Hardware and potential algorithmic weaknesses GPS-INS integration Magnetometer integration Baro-altimeter integration Foot-mounted INS and back-mounted PDR - type of movement (walking, running, crawling) and trajectory, - shoes, load being carried and surface, - quality of IMU (gyro bias drift, noise, dynamic range), Initialization of position, velocity and orientation

24 Needs for continued research from my perspective Multisensor system required Not too difficult to put together a fusion filter that works decently when sensors delivers good data However, this is not the typical condition, consider e.g. - GPS in urban environments and outdoor-to-indoor transitions - INS during irregular motions such as crawling, elevators, stomping - Magnetometers often encounter large local disturbances indoors - Barometric altimeters experiences anomalies due to heat and air turbulence, and other pressure changes The key challenge is then to try to have a good instant knowledge of how reliable the sensor data is Robust sensor fusion through sub-system integrity

25 Needs for continued research from my perspective Algorithms for reducing the sensitivity towards how the first responder moves Concerns foot- and back-mounted IMUs, and Doppler radars Camera-based positioning Reduced computational complexity Thermal sensors Improved sensors Especially better gyros and cheaper & smaller UWB-transceivers

26 Questions?