A WIFI/INS Indoor Pedestrian Navigation System Augmented by Context Feature

Transcription

1 A WIFI/INS Indoor Pedestrian Navigation System Augmented by Context Feature Ling Yang 1, Yong Li 2, Chris Rizos 3 Abstract. An Inertial navigation System (INS) is self-contained, immune to jamming/interference and in many other ways is ideally for pedestrian navigation applications especially in indoor environments. However, due to the sensor properties the quality of the navigation solution from a stand-alone INS will degrade rapidly, and must rely on some form of external correction/calibration/aiding to ensure system stability and reliability. In this contribution the authors use GPS when outdoors and WiFi when indoors to aid the INS in order to support seamless pedestrian navigation. To improve the performance of WiFi positioning, an enhanced fingerprinting method is proposed. A new fingerprinting database augmented by two types of map-based information is described. One is the topological relationship linking corridors, and the other is the orientation information of different corridors. Test results confirm that the navigation accuracy and stability is improved. Keywords: pedestrian navigation, step detection, walking status detection 1 Introduction A navigation system that tracks the location of a person is a useful capability for firefighters or other emergency first responders, for location-aware computing, personal navigation assistance, mobile 3D audio, augmented reality, and other applications. Such a system may be referred to as a Pedestrian Navigation System 1 Ling Yang ( ) School of Civil and Environmental Engineering, UNSW, Sydney, NSW 2052, Australia College of Surveying and Geo-informatics, Tongji University, Shanghai, China; ling.yang1@student.unsw.edu.au 2 Yong Li ( ) School of Civil and Environmental Engineering, UNSW, Sydney, Australia yong.li@unsw.edu.au 3 Chris Rizos ( ) School of Civil and Environmental Engineering, UNSW, Sydney, Australia c.rizos@ unsw.edu.au

2 2 L. Yang, Y. Li, C. Rizos (PNS). Global Navigation Satellite System (GNSS) is a well-known and extremely attractive technology for outdoor PNS applications. However GNSS-alone positioning may not satisfy the navigation integrity and reliability requirements due to its signal vulnerability to signal blocking or Radio Frequency Interference (RFI) (Ochieng et al. 2003). Integration of GNSS with a low-coast MEMS-based Inertial Navigation System (INS) has gained considerable attention for consumer-based navigation applications (Li et al. 2006, 2012; Godha et al. 2007, 2008). However, many pedestrian navigation applications involve walking from outdoors to an indoor destination point, which requires seamless navigation in both outdoor and indoor environments. Therefore, a GNSS/MEMS-INS integrated system is not entirely suitable since the navigation performance will severely degrade during long GNSS signal outages. Over many years different technological solutions have been investigated. Nowadays the most common alternative solution is based on wireless communications infrastructure such as WiFi. The preferred class of positioning algorithms makes use of the Received Signal Strength (RSS). There are two main types of RSS-based location determination techniques, the trilateration approach and the signal fingerprinting approach. The trilateration approach uses a signal propagation model to convert the RSS into a distance measurement, and several simultaneous distance measurements are then used to calculate the position of the WiFi tracking device (Akyildiz et al. 2009; Li et al. 2006; Chen & Kobayashi 2002). This technique is simple and comparatively easy to implement, however it suffers from the requirement for realistic signal propagation models. Moreover the exact locations of the WiFi access points are required. The second approach, known as fingerprinting, uses an empirical model to describe the distribution of RSS at various locations across the coverage area of interest (Evennou & Marx 2006). This method does account for the impact of the real environment on WiFi signal propagation. However, the main disadvantage is the need to create and maintain the RSS database across the area of interest (Li et al. 2005; Youssef et al. 2003). Recent investigations indicate that WiFi stand-alone positioning may not be able to satisfy the stability and reliability requirements for many mission-critical applications in busy or dynamic environments (e.g. due to temporary pedestrian signal obstruction). In this paper, in order to satisfy the accuracy requirement of indoor navigation, WiFi positioning technology is chosen as the most appropriate alternative for GNSS to integrate with INS. The advantages of using INS and WiFi for indoor positioning applications are their availability and relatively low cost. Two kinds of map-based information are introduced into the fingerprinting database to support the WiFi-only and WiFi/INS integrated solutions. One is the topological relationship between connected corridors. The other is the orientation information of different corridors, which is used as an attitude constraint in a WiFi/INS integrated system. With instruments mounted in a backpack, the GNSS/WiFi/INS integrated navigation system has been tested and its performance analysed. 2 System design and algorithm

3 Chapter Number Chapter Title ( Running head right style) 3 The Extended Kalman Filter (EKF) estimation algorithm is triggered at every GNSS/WiFi measurement epoch using the difference between the GNSS/WiFi and the INS mechanisation solution (loosely-coupled or tightly-coupled). The EKF algorithm can be found in (Hwang et al. 2005). The EKF estimates the INS errors using the measurement update equations. Whenever the GNSS/WiFi measurements are missing, the EKF operates in a time prediction mode, thus resulting in errors in position, velocity and attitude estimation that grow with time. [ ] { (1) [ ] where and are the navigation error vector and the inertial sensor measurement error vector, respectively. The psi-angle error equations of the INS are expressed in the navigation frame (or local east, north and up) for position, velocity and attitude updating (Goshen- Meskin & Bar-Itzhack 1992): { ( ) (2) where the superscript (n) refers to the navigation frame (n-frame), while the subscript (i) and (e) denote the inertial and the earth-centred earth-fixed frame (iframe and e-frame), respectively;, and are the position, velocity and attitude error vector, respectively; is the Earth rotation velocity; is the angular rotation velocity of the n-frame with respect to the i-frame; is the specific force vector; is the error of the gravity vector in the n-frame; is the rotation vector from the e-frame to the n-frame; is the transition matrix from body frame (b-frame) to the n-frame; and b and ε b are the accelerometer and gyro drift vector expressed in the b-frame, respectively. The dynamic matrix is obtained by a linearisation of Eq. (2). The measurement model is: ( ) (3) where x is the error state vector including and in Eq.(1); is the measurement vector; is the measurement variance-covariance matrix; is the design matrix that relates the measurements to the state vector; and is the measurement residual vector. 3 WiFi augmentation for position and orientation determination The WiFi fingerprinting algorithm consists of the following two steps: 1) In the training phase, the RSS measurement set from several Access Points (APs) at Reference Points (RPs) are measured and recorded as features of the RPs, to form a database for subsequent positioning. 2) In the positioning phase, the RSS measurement set observed at an unknown point is compared with the signal characteristics of the RPs contained within

4 4 L. Yang, Y. Li, C. Rizos the database. The RP signal data which have the closest match with the RSS measurement set are chosen as the estimated location. Since the position estimation actually is derived from the fingerprinting database, the quality of the database directly impacts on the final position accuracy. One option for improving accuracy is to build a database with denser set of sampled RPs. However this would require the investment of significant additional survey resources during the training phased, and would increase computation burden in the positioning phase. Another way to improve the quality of the database is to consider additional information that provides more constraints in the searching and matching algorithm. The next subsection describes a new database for WiFi positioning that incorporates map-based information. 3.1 Topology modelling in the fingerprinting database Two kinds of map-based information are introduced into the database to support the WiFi-only and WiFi/INS integrated positioning solutions. The information is derived from assumptions that pedestrian travel in indoor environment mainly occurs in corridors, for which two conditions apply: 1) The person cannot directly pass from one corridor to another disconnected corridor. 2) When walking in a corridor, the heading direction always coincides with the direction of the corridor. Figure. 1 Topological elements and relationships of a layer inside a building Noted that this study is concerned with the walking behaviour from an outdoor origin to an indoor destination, thus those actions such as pausing and facing a wall are not considered. Based on the above two assumptions, the topological relationship between any two corridors can be defined. Topology, which is a very common concept in GIS (Geographic Information System), expresses the spatial relationships between connecting or adjacent vector features (points, polylines and polygons). Corridors inside buildings can be abstracted as polylines, thus the road topology concept in GIS can be used to describe the relationships among corridors in an indoor area. In addition, the trajectory of walking along corridors can be

5 Chapter Number Chapter Title ( Running head right style) 5 modelled as a set of discrete points. Figure 1 shows the topological elements and relationships among corridors. Figure 1 graphically illustrates the two types of elements and their topological relationships. Labels 1 to 8 are called nodes, which are the connection points of adjacent corridors. Labels A to I are refered to as edges in a standard GIS database, and are corridors within a building. The arrows on the edges denote the direction of the edges, which are stored in the database as a feature of each edge. The topological relationships among these nodes and edges can be described as collections of geographic features, which are listed in Table 1. Table 1 Edge structure Edge Direction From Node To Node A B C D E F G H I Figure 2 Fingerprinting database structure Considering the above topological relationships, the RPs features stored in the fingerprinting database will also include this geographic information. Figure 2 shows the structure of the new database known as RP. The fingerprinting database actually stores features of a set of RPs, which are preselected points with known coordinates along the corridors. The coordinate of each RP is stored in the structure of LatLon. During the database building phase, a mobile user device is located at each RP for a short period of time. The MAC address and RSS measurements from each AP are collected and stored in the AP structure, named as MAC_ad and RSS. The new database should also contain a topology table that describes the geospatial information of each corridor, similar to Table 1. Depending on their particular topological properties these RPs are divided into two classes and associated with different features.

6 6 L. Yang, Y. Li, C. Rizos 1) For RPs which are very close to a connection point of several corridors, the corresponding node index will be recorded in the NodeID structure, and 0 will be recorded in the EdgeID. 2) For RPs which are distant from a connection point, the corresponding edge index will be recorded in the EdgeID structure, and 0 will be recorded in the NodeID. By searching the values of EdgeID or NodeID in the topology table, one can determine which corridor the user is located in/at, and can also obtain the orientation information. With this additional information the positioning accuracy can be improved. Geospatial topology is fundamentally used to ensure data quality as well as to aid data compilation. The use of topology in this study can be grouped into two aspects. The first is to augment the fingerprinting database for WiFi positioning. As described above, the topological relationships among each RP are stored together with the RSS measurements and coordinates, thus the features of RPs are extended. The other use is to provide constraints in the positioning phase, which will be discussed below. Firstly, in WiFi-only position the candidates which are not adjacent or connected with the previous position determination will be excluded. Then the orientation information associated with the selected candidates will be used as pseudo-measurement in the WiFi/INS integrated solution step. In the first step the accuracy of the prior position is important. An errorous prior estimate will bias the decision concerning the exclusion of candidate solutions. By integrating INS with WiFi, the short-term reliability and stability of the prior position can be guaranteed. 3.2 Searching and matching algorithms In fingerprinting positioning phase, an RSS measurement set is collected at each sampling epoch. By comparing the RSS measurement set with those stored in the database, a user s position can be obtained. Some techniques are deterministic, meaning that the applied RSS data are mean values and an RSS measurement set is matched with only one reference point. Transparent location fingerprinting uses a map of RPs (Brunato & Kiss Kalló 2002). A RP consists of a sequence of pairs (ss i, c i ). ss i is a RSS measurement set and c i is the corresponding physical coordinates. The matching of a new measurement set ss is done by selecting a number of RPs (k) that are closest to the measurement. c f, the weighted average of their coordinates, is returned: ( ) ( ) where d(ss j, ss) is the Euclidean distance between the two RSS measurement sets and ε is a small real constant. The probabilistic techniques use a RSS probability distribution at each RP rather than a mean value. Roos et al. (2002) use two estimators to match a measurement with the database. In the first method, a probability mass is assigned to a (4)

7 Chapter Number Chapter Title ( Running head right style) 7 kernel around each observation in the training data. The second method is closely related to discretization of continuous values. Without considering the geospatial relationships among these points, some faulty RPs which are physically far away from the mobile terminal may be selected. To exclude these incorrect RPs from the k candidates, the following three criteria based on the topological database are enforced 1) The candidate which is not adjacent with the prior location of the user will be excluded; 2) The candidate which is not consistent with the user s walking direction will be excluded; 3) The candidate which is not within the estimation error circle will be excluded. Figure 3 A demonstration of the exclusion based on topological information Figure 3 is a demonstration of the exclusion decision based on the above three criteria. The yellow star point is the prior estimated position based on the previous coordinates (red star point), and its estimated error is indicated by the circle. The rectangular points are k candidates obtained by the searching algorithm. With the first criterion, the two green rectangular points are excluded since they are actually located at corridor F which is not connective with corridor D, where the current location is supposed to be. Using the second criterion, the orange rectangular point is excluded. The other two blue rectangular points are also excluded by the third criterion. Finally the red rectangular point is obtained as the estimated position. The three criteria may not be always effective. The first criterion will not work for a single corridor. The second criterion is not relevant when the user is turning since the instantaneous walking direction is difficult to determine. The third criterion largely relies on the estimation error thus a miss-detection will occur if the error radius is large or a wrong exclusion will be introduced using an over-strict error radius. 4 Experiment and analysis 4.1 Experiment description

8 8 L. Yang, Y. Li, C. Rizos Two experiments were analysed to assess the navigation performance under different indoor environments. In these experiments the distance between two APs is an average of 1 m. The data is sampled at each RP for 1 minute. The nax5, a small MEMS-INS/GPS device, was used to collect inertial and GPS data, with the GPS antenna fixed to the top of the device as shown in Figure 4. For these tests the output rate of the inertial sensors was set to 10Hz. The WiFi data was collected by a laptop with built-in wireless internet card with a sample rate of 1Hz. Figure 4 Experiment setup Figure 5 Building layout for Experiment 1 The first experiment was conducted on the lower ground level of the Electrical Engineering Building at the University of New South Wales (UNSW), Sydney, Australia. The building layout is shown in Figure 5. The second experiment was conducted on level 2 of the Eastgardens Shopping Center, Sydney, Australia, and the building layout is shown in Figure 6. The red points in Figures 5 and 6 are the RPs in the corresponding fingerprinting databases. Figure 6 Building layout for Experiment Performance evaluation of WiFi-only system Experiments 1 and 2 are typical layouts of offices and shopping centres. In the office building rooms are usually located at two sides of the corridors. A common design of shopping centres locates shops around one side of the circular corridors.

9 north (m) north (m) Chapter Number Chapter Title ( Running head right style) 9 Figures 7 and 8 show the positioning results of these two experiments, where the blue marks are results by using traditional fingerprinting methods, and red marks are results obtained using the proposed method taking into account the topological information associated with corridors with topology information without topology information east (m) Figure 7 WiFi-only positioning results of Experiment with topology information without topology information east (m) Figure 8 WiFi-only positioning results of Experiment 2 It can be seen from Figure 7 that the WiFi positioning results with and without topological information are similar. In contrast, Figure 8 shows an obvious improvement by using topological information for Experiment 2, where the red points have a much smoother trajectory than the blue points. The different performances in these two experiments are likely due to the different characteristics of the building corridors. In Experiment 1, there are only two parallel corridors which are more than 30 m apart and connected by a perpendicular corridor. It is nearly impossible to miss-locate one point to the other corridor which is not connected with the current corridor. However, the building layout in Experiment 2 is more complex. There are seven corridors, and some of them are parallel but close to each other. As a result the possibility of locating a point in a wrong corridor is high. By considering the topological relationships among corridors, the probability of miss-matching can be reduced. 4.3 Performance evaluation of the WiFi/INS integrated system The position and attitude solutions for Experiment 1 are shown in Figures 9 and 10, where the red line is the GPS/INS solution, the blue and green lines are results from WiFi/INS integrated system, without and with map-based information. The

10 yaw (deg) north(m) 10 L. Yang, Y. Li, C. Rizos performance in Experiment 2 is similar and not presented here due to space constraints GPS/INS GPS/WiFi/INS GPS/WiFi/INS-topo start east (m) Figure 9 Trajectories of Experiment 1 (red:gps/ins, blue/green: GPS/WiFi/INS without/with topology constraints) The red line in Figure 9 shows that the INS-only solution diverges quickly in an indoor environment. The blue line indicates that upon adding position corrections from WiFi the divergence can be largely controlled, such that the trajectory is much more consistent with the geometry of the corridors. The green line indicates the best performance, when considering map-based information in the integrated system. This means that by using building orientation constraints, the position accuracy and solution stability is improved WiFi/INS WiFi/INS-topo time (0.1 sec) Figure 10 Yaw results of Experiment 1 during indoor period (red: reference from building orientation information, blue/green: results of WiFi/INS without/with topological constraints) Figure 10 shows the attitude solutions without and with topological constraints. The red lines are the references solution from the building orientation information. It can be seen that the green points are much more consistent with the references, while the yaw angle given by the blue points are not accurate and not stable even when the person is walking along a corridor. This further indicates the accuracy

11 Chapter Number Chapter Title ( Running head right style) 11 and stability of attitude estimation is improved by considering the building orientation information in the fingerprinting database. 5 Concluding remarks The extensive use of WiFi-based positioning provides an alternative to realise continuous navigation as a person moves from outdoors to indoor environments, where GPS becomes unavailable. In this paper an enhanced WiFi database is proposed that takes into account the topological relationships among corridors and orientation information of each corridor. By integrating WiFi with INS, the longterm navigation stability for indoor environments is further improved. Experiment results have verified that by adding building/corridor topological information into the fingerprinting database, incorrect candidate solutions can be excluded with a higher success rate, and the searching and matching reliability is improved in WiFi-only positioning. By using the more accurate position measurements as well as the orientation constraint provided by WiFi, the performance of a WiFi/INS integrated navigation system is also improved. The orientation constraint has a significant impact on accuracy of position and attitude estimation. Acknowledgement The first author was sponsored by the China Scholarship Council (CSC) for her PhD studies at the University of New South Wales, Australia. The authors would like to thank Dr Binghao Li and Jiantong Cheng for their assistance regarding the traditional WiFi positioning algorithm. 6 References Akyildiz I.F., Sun Z., Vuran M.C. (2009) Signal propagation techniques for wireless underground communication networks. Physical Communication, 2(3), Brunato M., Kiss Kalló C. (2002) Transparent location fingerprinting for wireless services. Proceedings of Med-Hoc-Net, Mediterranean Wokshop on Ad-hoc Networks, Baia Chia, Cagliari, 17 September Report No. DIT Chen Y., Kobayashi H. (2002) Signal strength based indoor geolocation. IEEE International Conference on Communications, New York, 28 April - 2 May, 2002, pp Evennou F., Marx F. (2006) Advanced integration of WiFi and inertial navigation systems for indoor mobile positioning. Eurasip Journal on Applied Signal Processing, 1, Godha S., Cannon M.E. (2007) GPS/MEMS INS integrated system for navigation in urban areas. GPS Solutions, 11(3), Godha S., Lachapelle G. (2008) Foot mounted inertial system for pedestrian navigation. Measurement Science and Technology, 19(7), (9pp).

12 12 L. Yang, Y. Li, C. Rizos Goshen-Meskin D., Bar-Itzhack I.Y. (1992) Unified approach to inertial navigation system error modelling. Journal of Guidance, Control, and Dynamics, 15(3), Hwang D.H., Oh S.H., Lee S.J., Park C., Rizos C. (2005) Design of a low-cost attitude determination GPS/INS integrated navigation system. GPS Solutions, 9(4), Li B., Salter J., Dempster A.G., Rizos C. (2006) Indoor positioning techniques based on wireless LAN. 1st IEEE International Conference on Wireless Broadband and Ultra Wideband Communications, Sydney, Australia, 3-16 March, 2006, paper 113, CD-ROM procs. Li B., Wang Y., Lee H.K., Dempster A.G., Rizos C. (2005) Method for yielding a database of location fingerprints in WLAN. IEE Proceedings- Communications, 152(5), Li Y., Efatmaneshnik M., Dempster A.G. (2012) Attitude determination by integration of MEMS inertial sensors and GPS for autonomous agriculture applications. GPS solutions, 16(1), Ochieng W.Y., Sauer K., Walsh D., Brodin G., Griffin S., Denney M. (2003) GPS integrity and potential impact on aviation safety. Journal of Navigation, 56(1), Roos T., Myllymäki P., Tirri H., Misikangas P., Sievänen J. (2002) A probabilistic approach to WLAN user location estimation. International Journal of Wireless Information Networks, 9(3), Suh Y.S., Park S. (2009) Pedestrian inertial navigation with gait phase detection assisted zero velocity updating. 4th International Conference on Autonomous Robots and Agents, Wellington, New Zealand, February, 2009, pp Youssef M.A., Agrawala A., Udaya Shankar A. (2003) WLAN location determination via clustering and probability distributions. 1st IEEE International Conference on Pervasive Computing and Communications, Fort Worth, USA, March, 2003, pp