Previous Abstract Return to Session D1 Next Abstract

Session D1: Robotic and Indoor Navigation

Utilization of Weak Received Signal Strength for Accurate Indoor Position Estimation
Toshiaki Yokoi, Department of Information Systems, Tokyo City University, Japan; Kazuki Oikawa, Digital Arts Inc., Japan
Location: Spyglass

According to the white paper from Fire and Disaster Management Agency of Japan in 2016, around 1200 people were killed on account of fire disaster in a year. More than half of the total were left behind and failed to escape, most lost their lives by carbon monoxide poisoning.
In contrast with outdoor navigating supports, the indoor localization is still an important subject of research. The indoor positioning estimation accompanies difficulty due to the complexity in electromagnetic wave environment and/or the additional cost for apparatuses which enable accurate location estimation. There are several approaches to identify the location by electro-magnetic wave fingerprints. The recent approaches uses the channel frequency response or the channel state, the combination of WLAN and BLE or WLAN and internal sensors, the only WLAN, etc. Also researches on the fast data collection, the database management and the data post-processing for accurate estimation are carried out. The cost problem is crucial to the socially disadvantaged people like children and aged people, etc. Through the several investigation for indoor navigation method to derive preferable way in both of cost and accuracy, we found the appropriate evaluation of the WLAN received signal strength in time-domain would provide the sufficient accuracy with reasonable cost. The proposed method uses only the ordinal smartphones and typical server for location estimation, but we adopted the mode value of the received signal strength to derive stable WLAN fingerprints. Our preliminary experiment showed sub-meter accuracy in location estimation. We also made a terminal software for administration office to recognize evacuees’ locations in a quasi-three-dimensional view using Java FX3D.
We have the typical access point allocation for a floor to guarantee the stable access to WLAN. The observed time variations of received signal strength for different intensities vary to time because of complicated electromagnetic environment. Especially, the detection probability of the very weak signal was around 70%. On the other hand, we can use the signals from other floors, which are relatively weak but continuously observed. The received signals from access points at plural floors provides abundant information for specifying the indoor location. However, the time variation of signals due to various cause seemed to make the estimation difficult. Since the time-average of the waving signal strength for a specific duration would vary in time, it cannot be used as stable indicator. Through the statistical analysis, we found the mode value is stable in most signals. We measured the received signal strength at specific points for 5 minutes with 2 second interval to derive mode values. Simultaneously, we recorded the SSID, BSSID, location data, floor number, date and time, etc. The fingerprint similarity among the current location fingerprint and stored fingerprint is evaluated by the summation of the squared difference in signal strength from same access points. When the current point is located among pre-observed points, the current location can be estimated by weighted interpolation of pre-observed point location. We derived the variation of the location estimation error to the number of pre-measured locations used for estimation at a specific point A. The result implies that the case of using all received signals yields the better accuracy than another. However, if we choose more than sixteen pre-measured locations used for estimation, the estimation error become bigger. The cause of this phenomenon is considered as the influence of very weak and discontinuous received signals. To avoid over-estimation of very weak and discontinuous signals, we should choose several pre-measured locations to be used in estimation from all. Our results implies that we should choose several observed signals from all in order to increase the estimation accuracy. As mentioned earlier, the reason of the increase in error for the use of many pre-measured location is considered as the influence of the very-weak and discontinuous signals. By choosing the appropriate number of pre-measured locations for estimation at specific points, we could derive accurate location. Here, we choose the number of fifteen as the number of the pre-measured locations for a point A, while choosing the number of three for a point B and the number of nine for a point C. The reduction in location error is achieved by the accurate evaluation of the mode value of RSS at the measuring point. Based on the result on the relation between the location estimation error and the number of pre-measured locations, we chose the number of six as the number of pre-measured locations in general location estimation at our site. From the location estimation procedure, we found that the error of location estimation at specific points of A, B and C become less than one meter after about 90 second estimation. Since the one meter accuracy is considered as less than human size, we are now able to identify the evacuee’s location. Although the 90 seconds standstill is long for moving person, it is acceptable for the person waiting the rescue. We believe that we will be able to shorten the required time for location estimation by improving the procedure to derive the mode values of RSS.
We also developed the Java-based location estimation system for a head office for disaster prevention. The system consists of the smartphone for WLAN fingerprint measurement, the server system for managing the WLAN finger prints and estimating the evacuees’ locations, and the administrator terminal for control. In order to rescue evacuees at emergency, the visualization of evacuees’ information is crucial to the head office. Hence we developed the evacuees’ location visualization software using the Java FX3D technology. This software enables the head office staffs to recognize current status of evacuees immediately. Each evacuee is identified by his/her handheld terminal ID.
Thus, this paper has described the method to derive stable fingerprint of the WLAN RSS for accurate indoor location estimation. By using the mode value of each received signal, we could derive sub-meter accuracy in location estimation. This accuracy is considered as sufficient for our application of this location estimation system for the rescue of evacuees at disasters. Based on the result, we also developed the fundamental visualization software for the head office for disaster prevention. We are now trying to shorten the location estimation time in order to realize the reliable and practical system for the socially disadvantaged communities.



Previous Abstract Return to Session D1 Next Abstract