Abstract²Smart phones and Location Based Services (L BSs) are getting very popular in recent years. To this end, locating the positions of smart phones becomes inevitable. To date, there exist some localization systems in some specific areas. For instance, GPS can be used only in the outdoor space. However, people always live in large scale environments like campus, urban and tourist areas, including both indoor and outdoor space. For large scale environment, a single location technique is not always available everywhere. Therefore, we propose a heterogeneous localization algorithm which combines GPS, WiFi and Internet Of Things (I O T) localizations. We also introduce the
³ORFDOL]DWLRQSRVVLELOLW\´IRUHDFKORFDOL]DWLRQPHWKRG to estimate the PRVW ³UHOLDEOH´ SRVVLEOH RQH Besides, the more reliable nearby users provide their localization information, the further enhancement of the localization accuracy will get. By measuring the relative WiFi signal strength, it helps especially those users who have no any available localization methods. We call this concept ³FRRSHUDWLYH ORFDOL]DWLRQ´ Finally, we evaluate the accuracy of our algorithms by simulation. Because signal strength fluctuates from minute to minute, we measure empirical data and put into the simulator to do our experiments. Finally, we also verify that our idea can be implemented on smart phones and work feasibly.
Keywords: localization; Wi F i; GPS; heterogeneous;
cooperative; large scale
I. INTRODUCTION
Personal Digital Assistants (PDA) and smart phones have become very powerful, e.g. Android and iphones. These devices always provide wireless Internet capabilities and Global Positioning System (GPS). These facilitate a novel classes of applications, called Localization based Services (LBSs). LBSs provide added values by enabling services such as resource tracking, finding someone or something, and proximity-based notification (push or pull). So there is a need for a feasible localization method any time anywhere.
For outdoor localization systems, GPS is easy to use and offers a good localization accuracy and reliability. But this technique needs at least three line-of-sight GPS satellites. So it is inefficient in indoor environments and also degrades the performance in urban environments. Moreover, the energy consumption by GPS is significant, leading to be not always on unless needed.
Indoor localization technique has become more popular in recent years. It can be used for the location detection of products stored in a warehouse, medical personnel or
equipment in a hospital, or firemen in a building on fire. Radio Frequency Identification (RFID) and WiFi localization can be used in indoor space. RFID can provide good accurate localization. SpotON[1] and LANDMARC[2] are RFID localization systems. But it requires specialized hardware like RFID readers and tags. WiFi localization methods mostly belong to location fingerprinting methods [3,4,5,6,7,8,9,10], using existing wireless access points to measure their signal strength maps. Many places currently are WiFi available, and smart phones are WiFi enabled. In this category of localization, Microsoft RADAR[3] proposed the first WiFi localization algorithm. However, this kind of methods is time consuming for building the fingerprint database, and needs a dense number of nearby wireless access points. Therefore, relying only on WiFi localization is not practical for positioning systems in large outdoor space.
There are indoor and outdoor spaces in urban environments or on campus. Most people are always staying, walking or biking in this environment. But GPS, WiFi and RFID localization have some drawbacks in indoor and outdoor environments, individually. Therefore, we propose a heterogeneous localization using combined WiFi, RFID, and GPS for localization in the large scale environments. We can imagine in the near future that there are many RFID readers in urban environments and on campus, like MRT payment systems and RFID access control systems on campus. People can use RFID and Near Field Communication (NFC) devices WRJHWUHDGHU¶VORFDWLRQV. Thus, a newly developed smart phone has all the three capabilities already. Furthermore we introduce the concept of ³cooperative localization´. If people keep WiFi on for their smart phones, they can scan nearby user¶s relative WiFi signal strength to determine their approximate distance.
In this case, even they don¶t have any GPS or RFID systems to use, they still can ask the nearby user¶s location to estimate their ³possible´ location.
The rest of this paper is organized as follows. Section 2 introduces related works in WiFi localization systems and other localization methods. Section 3 introduces the basic idea and algorithm of our localization algorithm. Section 4 presents simulation results and the implementation experiments. Finally, Section 5 concludes this paper.
II. RELATEDWORK
There are a lot of wireless-based localization systems in recent years. But they are only used in some specific areas.
UWB can be used in indoor and outdoor environments but it is applicable only for line-of-sight environments. RF and IR, RF
and ultrasonic hybrid methods are used only in indoor environments. And they need specialized hardware. WLAN localizations perform well in indoor and restricted outdoor environments. Therefore, we use GPS and WLAN to propose our heterogeneous localization in large scale environments.
We first introduce some exiting WiFi based localization systems and analyze their drawback in the large scale environments.
Two localization algorithms are presented, namely Collaborative localization [11] and Virtual Compass[12].
They use nearby user¶s location to correct the localization.
The Collaborative localization method IRXQG FRQ¿GHQFH LQ
location estimation to compute the positioning stability of a target node ovHUWLPHIURPDSDUWLFOH¿OWHU %\¿QGLQJQHDUE\
targets with proximity sensors (Zigbee), nodes with lower FRQ¿GHQFH FRXOG LPSURYH WKHLU HVWLPDWLRQ DFFXUDF\ E\
leveraging more accurate location information from nearby neighbors within the same cluster. The Collaborative localization has two drawbacks in large scale environments.
First, they use sensor to detect nearby people within 2 meters.
We think 2 meters is too short for large scale environments.
Second, the system is only WiFi localization system. It can be used only in indoor and some restricted outdoor environments.
Virtual Compass is a peer-based localization system for mobile phones. It does not require any infrastructure support, but uses multiple, common radio technologies instead, like WiFi and Bluetooth to create a neighbor graph. Virtual Compass is focused on relative positioning, because mobile social applications are heavily driven by the relative FDQ¶W XVH RQO\ RQH ORFDOL]DWLRQ DOJRULWKP. Therefore, in this section, first we introduce WiFi localization system using probabilistic approach. Second, we propose the heterogeneous localization. Third, furthermore we consider nearby neighbors to propose the cooperative localization.
A. WiFi Localization
Our previous work with Service Science Research Center (SSRC) in NCCU, we developed a WiFi localization system at Taipei World Trade Center Nangang Exhibition Hall to provide smart navigation for foreign buyers. And we tested our localization system in 2010 Taipei Int'l Electronics Show, 2011 Taipei Int'l Machine Tool Show and Cycle show. The average accuracy is 4.2 meter in Taipei Cycle show. Figure 1 shows the experimental results in the Cycle show.
Figure 1: Cumulative percentage of error distance in Cycle show.
The following introduce the WiFi localization method we used in the system.
1) Data Collection:
There are 56 Aps, spacing at 17 meters apart in the exhibition hall. During the offline phase, we measured signal strengths every 6~8 meters, with 20 measures at one point by Android devices. These data are stored for database.
2) Radio Map Creation:
Our system loads those raw data at one point in database and calculates mean and standard deviation of every AP. This is called radio map and the point in radio map called radio map point.
3) Tracking Data Collection:
During the online phase, the user takes the Android device to measure signal strength at each tracking point. Every tracking points measure signal strength five times and the system calculate mean and standard deviation of every AP.
4) Target Determination:
The system calculated the similarity[7] between tracking point and every radio map points to search five closest match radio map points. These five points called targets. The algorithm creates two signal strength normal distributions to tracking point and radio map for one access point and computes the intersection area of the two density distributions.
Overlap(TP, RMP) = (1)
Similarity(TP, RMP) = /m (2) Equation 1 calculates the overlap area between tracking point and a radio map point of one access point. TP is tracking point. RMP is one of radio map point. Equation 2 is the average overlap of all access points between tracking point and one radio map point. n is the set of common access points.
m is the number of all access points measurable in both points.
5) Interpolation and Location Estimation:
We use five target points and their weight to interpolate our location estimation. The weight of every target points is the inverse of Euclidean distance between tracking point and the target point in signal space. And then we calculate weighted sum by targets¶ coordinate and weight.
6) Filtering:
Signal strength fluctuations from minute to minute, and the fluctuations arise also from motion of people. It might cause the predicted location to be far away from the previous location. Therefore we use filter to correct the location estimation like Collaborative localization [12].
B. Heterogeneous Localization
We use WiFi localization in Nangang exhibition hall, because there are a lot of APs. But in this paper we focus on the large scale environment, and AP is sparse in more outdoor space. On the other hand, there could be a lot of RFID readers in urban environments and on campus. So, we can use NFC or RFID tags in the smart phone to get the location from those readers. Therefore, we proposed a heterogeneous localization algorithm to combine GPS, IOT and WiFi localizations.
Think about which location is accuracy and reliability, if we can get two or three locations at the same time? We use
³localization possibility´ to solve this problem. The higher possibility means the localization is possible to be more accurate. Our method gives IOT localization possibility to be 1, because the communication distance of the IOT reader and the tag is very short. If the GPS device can get its localization, we set the possibility to be 0.35 by default. We will explain this later in the simulation. It is reasonable that the GPS localization possibility is not 1, because GPS is not always accurate in the outdoor space. WiFi possibility is set to be the similarity of the first target point in WiFi localization algorithm. Localization possibility is reset to 0 when the ORFDOL]DWLRQ PHWKRG FDQ¶W HVWLPDWH any location. The localization possibility is used to enhance the particle filter in Collaborative localization[12]. The filter corrects the location when the location estimation exceeds the human movement capacity. But if the previous location estimation is not so accurate, then the new location is bounded in the human movement capacity. This will cause error localization estimation. So our method compares the possibilities between the previous location estimation and the new location estimation. If the possibility of new location estimation is higher than the previous one, then the location estimation can
³jump´, namely using new location to filter.
Figure 2 shows an example for the heterogeneous localization. Suppose a user is estimated on Lt by localization system at time t. And then the user moves. The localization system also uses GPS WiFi and IOT localization. And they have their location possibilities. WiFi localization possibility is the similarity of first target in WiFi localization. IOT¶s possibility is 1 when it gets reader¶s location. We select the coordinate which have highest possibility at time t+1 and the coordinate is t+1. At the same time, the possibility of the original localization should be reduced, because the user moves from time t to t+1. Pt¶ is the reduced localization possibility from Pt, say 0.9 in this example. At the same time, we also calculate the moving speed by localization history.
Assume the system refers previous 4 localizations and calculates the average moving speed. Therefore, we get human movement capacity by multiplying moving speed and time interval from time t to t+1. We can draw two human
movement capacities around Lt and t+1. If there are intersection with the two human movement capacities and the localization estimation is on the middle of them. Lt+1 is the localization result. And Pt+1 inherits the higher possibility of Pt¶ and t+1.
Figure 2: An example of heterogeneous localization.
If there is no intersection of the two human movement capacities like in figure 3. In case A, if Pt¶ is higher than t+1
and we use filter to correct the location estimation to human movement capacity around Lt. In case B, if Pt¶ is lower than
t+1 and we think t+1 is more accurate. So location estimation is on the human movement capacity around t+1.
Figure 3: Two human movement capacities are not intersected.
Case A is using filter.
Case B is jumping to new location.
In the heterogeneous localization, we used localization possibility to select a localization method from GPS, WiFi and IOT localization. We do not use three localization coordinates simultaneously. Because we think some localization is not accurate. Localization accuracy will be degraded a lot if the inaccurate localization is used to estimate.
In the heterogeneous localization, we reduce localization possibility of the previous localization estimation as time passes. Figure 4 shows how we reduce the possibility. When the user moves to Lt+1 from Lt at time t+1, the possibility should be reduced if we use Lt for Lt+1. We refer the localization estimation from Lt-3 to Lt. First we calculate mean
Case A.
Case B.
coordinate Lmean of Lt, Lt-1, Lt-2 and Lt-3. Second, we calculate the distance dt, dt-1, dt-2 and dt-3 from Lt, Lt-1, Lt-2 and Lt-3 to Lmean. Third, we calculate the variance of dt, dt-1, dt-2 and dt-3. Forth, we use equation 3 to calculate possibility reduction.
Here, we consider variance of previous localization and elapsed time interval. Pt+1 is localization possibility at time t+1. Pt is previous localization possibility. is variance of dt, dt-1, dt-2 and dt-3. is time interval from t to t+1. D and T are constant. Moving variance beyond D or elapsed time exceeding T, means the previous location has no longer value for localization. The possibility is 0<P<=1, so we use minima with 1 and maxima with 0 in equation 3.
(3)
Figure 4: An example for possibility reduction.
C. Cooperative Localization
So far, we have introduced a heterogeneous localization algorithm to combine GPS, WiFi and IOT localization in large scale environments. And furthermore if neither GPS, WiFi or IOT localization is not available, how do we get my location?
Here, we will FRQVLGHUQHDUE\QHLJKERUV¶UHODWLYH:L)LVLJQDO
strength and location, and propose our cooperative localization method.
Figure 5 shows an example for the cooperative localization.
Assume that N0 wants to estimate or correct its location by nearby node N1 and N2. Assuming that N1 and N2 have higher possibilities than N0. They can scan the relative WiFi signal strength, then translate N1¶s RSSIN0 and N2¶s RSSIN0 to distance D1 and D2 by equation 5. Equation 5 is proposed by RADAR [5].
(4)
n indicates the rate at which the path loss increases with distance. P(d0) is the signal power at some reference distance.
d is the transmitter-receiver separation distance. C is constant.
And we calculate the possible location LN0(N1) within D1
from LN1 to LN0. We also calculate possible location LN0(N2)
within D2 from LN2 to LN0. Finally we calculate weighted sum by LN0, LN0(N1) and LN0(N2) and PN0, PN1 and PN2. So LN0c is the resulting location and PN0c is the highest possibility of PN0, PN1
and PN2.
Figure 5: An example for cooperative localization.
IV. EXPERIMENTAL EVALUATION
A. Simulation setup
To date, PDQ\ORFDOL]DWLRQV\VWHPVGLGQ¶WXVHVLPXODWRUWR
test their performance, because signal strength fluctuations from minute to minute. But we measure the actual signal for our simulator to easily do a lot of estimates during the node moving. The simulator was implemented in Java and database is SQLite.
First, we use HTC Desire(Android 2.2) to site survey the 1/4 our down-hill campus every 5~10 meters. Figure 8 is an example for site survey. Every point contains WiFi signal strength, longitude and latitude. Second, we take the HTC Desire and walk fifteen paths on campus and the Desire records the actual longitude and latitude, WiFi signal strength and GPS longitude and latitude about every 1 second.
Figure 8: Left is an example of site survey.
Right is an example of path.
There is a problem that how to simulate relative WiFi signal strength between users? Therefore, third we measure WiFi signal strength every five meters using 2 HTC Desire before the simulation. We save this to database. Forth the simulator creates and run fifteen path threads. Fifth, the simulator calculates the distance between every pair of path by their actual longitude and latitude. Sixth, the simulator matches the distance to relative signal strength between them. Figure 9 is the distance to WiFi signal strength in the outdoor environment between two HTC Desires . Red line (Meter to RSSI) is the actual WiFi signal strength between two Desire
mobiles and blue line (formula: RSSI to Meter) is from equation 4. We take every pair of point on red line (Meter to RSSI) to equation 4 and calculate n=2.94. Therefore the red and blue lines are closest matched when n=2.94 in figure 9. In the simulation we use figure 9 in our simulator. And we use relative WiFi signal strength when distance is 25 meters between two Desire mobiles.
Figure 9: Distance (Meter) to WiFi signal strength (RSSI) in outdoor space. X axis is distance between two Desire and Y axis is relative WiFi signal strength.
Seventh, the simulator runs GPS, WiFi, heterogeneous and cooperative localization. Finally, the simulator calculates the error distance between the actual location and the localization estimation of GPS, WiFi, heterogeneous and cooperative localizations.
Figure 10: WiFi localization possibility and error distance in all paths.
Figure 10 shows the WiFi localization possibility and error distance in all paths. In this figure, the y axis is WiFi localization possibility and the x axis is error distance of WiFi localization. We set GPS possibility to be 0.35, because there are just a few points with error distance more than 20 meters when possibility is higher than 0.35. Thus, when we use both GPS and WiFi localization simultaneously, our proposed heterogeneous and cooperative localization will select GPS when WiFi localization possibility is lower than 0.35.
B. Simulation result
We measure fifteen paths for the simulation. We will show four paths localization results in this section. We use cumulative percentage of error distance graph to show the
results. X axis is localization error distance and Y axis is cumulative percentage of error distance (%).
Figure 11 shows the simulation result for outdoor path (path 1). In this case, GPS is more accurate than WiFi localization. And our heterogeneous localization indeed select GPS more often. Therefore, the heterogeneous localization and GPS have similar accuracy. Furthermore, if there are two paths nearby path1, they can correct the localization error of path1 by the cooperative localization.
Figure 11: Simulation result in path 1
Figure 12 shows the simulation result for the path from indoor to outdoor to indoor (path 2). This path passes indoor space, so GPS performs poorly in part of the path. We can see from the results that our heterogeneous localization selects GPS in outdoor environments and selects WiFi in indoor environments. The heterogeneous localization is more accurate than only WiFi localization. Furthermore, the cooperative localization corrects the localization by nearby paths.
Figure 12: Simulation result in path 2
Figure 13 shows the result for indoor path (path 3). Again, GPS performs poorly in indoor environments. WiFi localization is very accurate in this building. The heterogeneous localization uses human movement capacity and localization possibility to correct localization of WiFi
Figure 13 shows the result for indoor path (path 3). Again, GPS performs poorly in indoor environments. WiFi localization is very accurate in this building. The heterogeneous localization uses human movement capacity and localization possibility to correct localization of WiFi