Indoor/ Outdoor Pedestrian Navigation with an Embedded GPS/ RFID/ Self- contained Sensor System Masakatsu Kourogi, Nobuchika Sakata, Takashi Okuma and Takeshi Kurata National Institute of Advanced Industrial Science and Technology (AIST)
Research background (1) � Location/ direction-based Web services are widely available to provide local information such as maps, weather and nearby transportation. (ex. Google Maps and Yahoo! Maps) � Portable PCs and PDAs (smart phones) are capable of rendering 3D urban landscape. (ex. Google Earth and Pocket Cortona)
Research background (2) � Combination of location/ direction- based services and a suite of 3D mapping software will provide highly intelligent navigation system. � It is essentially important to acquire accurate position and direction to enable such navigation system.
Research targets � In both indoor/ outdoor environments, to achieve stride-level accuracy of positioning method. � It is realized by dead-reckoning method combined with RFID and GPS. � To be implemented by an embedded computing system and provide pedestrian navigation services.
Proposed method: Dead-reckoning based on walking locomotion � Self-contained sensors (gyroscope, magnetometers and accelerometers) realize dead-reckoning based on human walking locomotion. � Partially proposed by our previous researches. � Dead-reckoning will work in both indoor/ outdoor environments.
Proposed method: Dead-reckoning based on walking locomotion � Accuracy of dead-reckoning is vulnerable to accumulation of step- wise error and thus requires external sources of information about absolute position to correct such error. � First, error model of dead-reckoning is required. � Second, GPS and active RFID tag system are used as external position correction.
Proposed method: Error model of dead-reckoning � The error of dead-reckoning is composition of that of azimuth and of stride. error The estimated azimuth The true azimuth The true stride The estimated stride The previous position
Proposed method: Error model of dead-reckoning � Combination of error caused by azimuth and stride are approximated by Gaussian distribution. θ Distribution for azimuth i Known to be Gaussian approximation Next position 2-D Gaussian distribution Distribution for stride l i Known to be Gaussian 1-step of dead-reckoning Distribution of the previous position
Proposed method: Error model of dead-reckoning � Kalman filter is used to update the estimation of position and velocity. � The state vector: [ ] = T s x y v v t t t x y t t Estimated from acceleration position velocity during the walking locomotion � Update equations: O Observation of the state t = + − ( ) s s K O s K Kalman gain + 1 | t t t t t t t P Covariance matrix − = + 1 ( ) t K P P R of the state vector t t t t = − R Covariance matrix of P P K P t + | 1 t t t t t the error of observation
Proposed method: Estimation of pedestrian velocity � The velocity and acceleration gap (amplitude) are highly correlated and estimated from the other. Subject A 8 Subject B Fitting line (A) 7 Fitting line (B) 6 5 Speed [Km/h] 4 3 2 1 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Forward acceleration gap [G]
Proposed method: Combination with external sources (1) � GPS is used in outdoor environment. � Error of GPS has three components: � Error caused by multipath effect � Offset error caused by signal delays. � Random error (represented by Gaussian) � Multipath error can be reduced with combination of dead-reckoning. 95% confidence area GPS data out of 95% area Estimated by dead-reckoning is excluded for computation The previous position by dead-reckoning GPS data included for computation
Proposed method: Combination with external sources (2) � GPS is used in outdoor environment. � Offset error can be measured by the fixed observation station whose location is exactly known. � Thus, remaining random error can be handled within Kalman filter framework since it has normal distribution. � Two measurements by GPS hints the position and velocity in the state vector.
Proposed method: Combination with external sources (3) � Active RFID tag system is used to correct the pedestrian position. � Error of position can be represented by the Gaussian distribution. Reachable range of the ID signal The height of user’s waist where the RFID tag is attached 1.3m 0.6m Range of position Connected via 1.5m wireless network Floor A RFID tag reader
Implementation: Outlook of the system � The total system is Orientation Tracker implemented in an embedded HMD Camera computing system. Camera � A browsing system is Orientation Tracker separately implemented. HMD user HMD user Handheld PC � HMD system and handheld system are implemented. GPS RFID Tag Self-contained sensors Embedded Computer Handheld display user Handheld display user
Implementation: Diagram of the system Browsing system (SONY VAIO type U) From/To Attitude tracker the control server Position/direction Data distribution of other users engine CCD camera Google Earth (Navigation application) Equipped with browsing PC JPEG image Adjustment Estimation results of position and direction request (in NMEA0183 or CSV format) Self-contained Dead-reckoning sensors Positioning GPS positioning GPS module (via Wireless network) engine From RFID positioning RFID tag the RFID tag system (via Wireless network) Equipped with Embedded computing system user’s hip
Implementation: Diagram of the system Dynamically generated KML data Google Earth CGI code Location/direction Query Web server JavaScript CGI code (Asynchronous) Query with location/direction CGI code Location/ direction Nearby contents Remote Web server Embedded system Navigation browsing system (with SQL database)
Our system in actions (Indoors) Demo video:
Our system in actions (Outdoors) Demo video:
Experiments: The ground truth vs estimation � Setting for experiment: � 368.1 meter route (outdoor: 247.2 meter, indoor: 120.9 meter) � Two RFID readers are placed in the building to correct user’s position. � Five subjects traveled along with the same route. � Estimations by the proposed method were compared to the ground truth of the route.
O u r m e t h o d Ground truth Due to drift errors in gyrosensors, the result of DR only dead-reckoning deviates. DR only Ground truth GPS only Our method Differentiation of GPS results GPS only adjusted the error in direction. 25m Start here
Experiments: The ground truth vs estimation � Error graph along with walking distance 30 Outdoor Indoor GPS only 25 DR only Averaged error (m) GPS results are used to Our method 20 adjust errors in direction/position 15 RFID adjustments 10 Elevator detection adjustment 5 GPS results are discarded 0 0 50 100 150 200 250 300 350 400 Walking distance (m)
Practical usage of the system: Openhouse situation with kids � 23 kids have experienced our prototype system. � No prior training or calibration is required. � User’s height is only parameter required by the system. � The system worked well even if kids moved in unexpected manners.
Conclusion � GPS/ RFID/ Dead-reckoning integrated positioning method is proposed. � Embedded pedestrian navigation system is implemented with the proposed method. � Accuracy of the proposed method is shown to be 5-10% of total walking distance. � Kids can play with the system in the open house situation.
Recommend
More recommend
