기존의 와이파이 기반 측위 시스템에서 주로 사용되는 칼만필터와 파티클 필터는 실내공간의 구조적 특성을 반영하지 못해 정확도가 낮고, 계산 부하 또한 높기 때문에 휴대기기를 이용한 실내 측위에 적용하는데 한계를 지닌다. 이러한 한계를 극복하고자 본 논문은 와이파이 기반 측위 시스템을 위한 적응형 혼합필터를 제안한다. 제안된 필터는 칼만 필터의 일반적인 적용 체계를 활용하였으며, 적은 수의 파티클을 사용한 파티클 필터의 개념 또한 추가되었다. 제안된 필터는 일반 칼만 필터와는 달리 예측 가중치를 동적으로 변화시켜 동작하며, 위치 예측을 위한 파티클을 실내공간의 경로 네트워크상에 한정하는 특징을 지닌다. 검증결과 적응형 혼합 필터는 일반 칼만 필터에 비해 높은 정확도를 보이며, 일반 파티클 필터에 비해서도 정확도 및 계산시간의 측면에서 유의할만한 성능향상을 보였다.
기존의 와이파이 기반 측위 시스템에서 주로 사용되는 칼만필터와 파티클 필터는 실내공간의 구조적 특성을 반영하지 못해 정확도가 낮고, 계산 부하 또한 높기 때문에 휴대기기를 이용한 실내 측위에 적용하는데 한계를 지닌다. 이러한 한계를 극복하고자 본 논문은 와이파이 기반 측위 시스템을 위한 적응형 혼합필터를 제안한다. 제안된 필터는 칼만 필터의 일반적인 적용 체계를 활용하였으며, 적은 수의 파티클을 사용한 파티클 필터의 개념 또한 추가되었다. 제안된 필터는 일반 칼만 필터와는 달리 예측 가중치를 동적으로 변화시켜 동작하며, 위치 예측을 위한 파티클을 실내공간의 경로 네트워크상에 한정하는 특징을 지닌다. 검증결과 적응형 혼합 필터는 일반 칼만 필터에 비해 높은 정확도를 보이며, 일반 파티클 필터에 비해서도 정확도 및 계산시간의 측면에서 유의할만한 성능향상을 보였다.
As the basic Kalman filter is limited to be used for indoor navigation, and particle filters incur serious computational overhead, especially in mobile devices, we propose an adaptive hybrid filter for WiFi-based indoor positioning systems. The hybrid filter utilizes the same prediction framework of...
As the basic Kalman filter is limited to be used for indoor navigation, and particle filters incur serious computational overhead, especially in mobile devices, we propose an adaptive hybrid filter for WiFi-based indoor positioning systems. The hybrid filter utilizes the same prediction framework of the basic Kalman filter, and it adopts the notion of particle filters only using a small number of particles. Restricting the predicts of a moving object to a small number of particles on a way network and substituting a dynamic weighting scheme for Kalman gain are the key features of the filter. The adaptive hybrid filter showed significantly better accuracy than the basic Kalman filter did, and it showed greatly improved performance in processing time and slightly better accuracy compared with a particle filter.
As the basic Kalman filter is limited to be used for indoor navigation, and particle filters incur serious computational overhead, especially in mobile devices, we propose an adaptive hybrid filter for WiFi-based indoor positioning systems. The hybrid filter utilizes the same prediction framework of the basic Kalman filter, and it adopts the notion of particle filters only using a small number of particles. Restricting the predicts of a moving object to a small number of particles on a way network and substituting a dynamic weighting scheme for Kalman gain are the key features of the filter. The adaptive hybrid filter showed significantly better accuracy than the basic Kalman filter did, and it showed greatly improved performance in processing time and slightly better accuracy compared with a particle filter.
* AI 자동 식별 결과로 적합하지 않은 문장이 있을 수 있으니, 이용에 유의하시기 바랍니다.
제안 방법
To track the orientation, this study applied additional Kalman filter to keep track of orientation. As the angular rate from gyroscope is cumulated, the error from gyroscope is also cumulated, so this study proposed a mechanism to periodically reset the sensor. Jirawimut et al.
In order to confirm the effect by the improved accuracy, we generated traces wi/wo applying AHF and then visualized the traces on the maps. As illustrated in Figure 3 the average error distance was a key factor for a generated trace to be close to the real trace.
In order to figure out how much restricting the prediction locations to the way network and dynamic weighting scheme contribute to the accuracy improvement, the accuracies were measured with/without incorporating each method. When we analyzed the results, in the improvement, the contributions of restricting the prediction locations to the way network was 7.
The lower dotted line shows the change of the true error distances and the upper line shows the change of the weight values. The experiment was performed with a trace sequence. As depicted in Figure 2, when the true error distance was long, a relatively low weight was assigned to the measurement, whereas if the true error distance was short, a relatively high weight was assigned to the measurement.
AHF achieved considerable improvement in the accuracy for both data sets of KAIST and of E-mart. The improvement was due to the restriction of the movement of the target object to the way network and the dynamic weighting scheme based on error estimation. We can expect further improvement if we develop a more advanced and reliable error estimation technique, which is an open problem.
To measure the accuracy of each method, we implemented a WiFi fingerprint-based localization engine using weighted kNN algorithm. Then BKF, AHF, and a particle filter were developed and integrated with the localization engine for the measurement. The particle filter conducted its prediction with 5,000 particles.
As gyroscope provides only angular rate, the filter estimates the orientation with cumulated angular rate. To track the orientation, this study applied additional Kalman filter to keep track of orientation. As the angular rate from gyroscope is cumulated, the error from gyroscope is also cumulated, so this study proposed a mechanism to periodically reset the sensor.
대상 데이터
The learning data was collected at 725 points, and 14,500 fingerprints were collected in total. For the test data, like the case of the KAIST library, 20 traces were collected, and each of the traces included 188 measure points.
In this paper, the evaluation compared AHF, BKF, and a particle filter performed on two data sets collected from KAIST library, Daejeon, Korea, and E-mart discount store, Seoungsu, Seoul. From the evaluation, the comparison results showed that AHF achieved accuracy improvements of 18.
E-mart is one of the biggest discount stores with 200 chains in Korea, and E-mart at Seongsu is the headquarters and main store of the chain. The data was collected on the 1st floor, which was 8800 m2. E-mart is also an open space, and the area is divided by dozens of shelves.
The evaluation compared the accuracy improvements achieved by AHF, a particle filter and BKF at the KAIST library and E-mart respectively. To measure the accuracy of each method, we implemented a WiFi fingerprint-based localization engine using weighted kNN algorithm.
Figure 3 shows the correlation between the true error distance and the error distance estimated by BCS method. The experiment was performed in N5 building of KAIST. Although the result is somewhat dependent on datasets, the overall correlation stayed in between 0.
The 4th floor of KAIST library is an open space of 2232 m2, and the space is divided by 67 bookshelves, making the way network complicated. The learning data was collected at 557 points, and 20 fingerprints were collected at each point. For the test data, 20 traces were collected, and each of the traces included 174 measure points.
E-mart is also an open space, and the area is divided by dozens of shelves. The learning data was collected at 725 points, and 14,500 fingerprints were collected in total. For the test data, like the case of the KAIST library, 20 traces were collected, and each of the traces included 188 measure points.
The validity of AHF for indoor navigation was evaluated on two data sets: one collected from the 4th floor of a KAIST library, KAIST, Daejeon, Korea and one from the 1st floor of E-mart at Seongsu, Seoul, Korea. The 4th floor of KAIST library is an open space of 2232 m2, and the space is divided by 67 bookshelves, making the way network complicated.
이론/모형
Particle filters are another approach to enhance the tracking accuracy by introducing numerous particles in prediction. The particle filters scatter numerous particles at a target area and use a Monte Carlo method to represent posterior distribution of the particles [5]. The location of the particles gradually changes according to received signals.
The evaluation compared the accuracy improvements achieved by AHF, a particle filter and BKF at the KAIST library and E-mart respectively. To measure the accuracy of each method, we implemented a WiFi fingerprint-based localization engine using weighted kNN algorithm. Then BKF, AHF, and a particle filter were developed and integrated with the localization engine for the measurement.
성능/효과
9% and the particle filter KK %. At the E-mart, as much as 25.0% accuracy improvement was made by AHF (from 4.88m to 3.66m), whereas BKF achieved only a 15.0% accuracy improvement and the particle filter 23.4 %.
In this paper, the evaluation compared AHF, BKF, and a particle filter performed on two data sets collected from KAIST library, Daejeon, Korea, and E-mart discount store, Seoungsu, Seoul. From the evaluation, the comparison results showed that AHF achieved accuracy improvements of 18.0 % at the KAIST library and 25.0 % at the E-mart discount store. AHF achieved slightly better accuracy improvement than the particle filter, but it showed greatly improved performance in processing time compared with the particle filter.
Usually, we can expect a greater filtering effect on straight lines than at corners or intersections. The accuracy results of AHF was around 10% better than what was achieved by BKF and 2 - 6% better than the particle filter.
In order to figure out how much restricting the prediction locations to the way network and dynamic weighting scheme contribute to the accuracy improvement, the accuracies were measured with/without incorporating each method. When we analyzed the results, in the improvement, the contributions of restricting the prediction locations to the way network was 7.93%, and dynamic weighting scheme was 3.23% at E-mart and 9.45% and 2.66%, respectively, at the KAIST library. The effect of dynamic weighting scheme was also observed in the particle filter because the particle filter using dynamic weighting scheme showed slightly better accuracy than the one using static weighting scheme.
후속연구
Signal filters and map matching are the typical filters that should be integrated together with AHF. The interrelations between the filters should be studied further, and analyzing the effect of applying AHF with the filters is our future work.
참고문헌 (16)
R. KALMAN, "A new approach to linear filtering and prediction problems", ASME DC Journal of Basic Engineering, vol. 82, no. 1, pp.35-45, Mar, 1960.
N. Sheimy, E. Shin, X. Jiniu, "Kalman filter Face-Off: extended vs. unscented Kalman filters for integrated GPS and MEMS inertial", InsideGNSS, vol. 1, no. 2, pp.48-54, Mar. 2006.
Y. Chiou, C. Wang, S. Yeh, "An adaptive location estimator using tracking algorithms for indoor WLANs", ACM Journals on Wireless Networks, vol. 16, no. 7, pp.1987-2012, Oct. 2010.
A. Doucet, S. Godsill, C. Andrieu, "On sequential monte carlo sampling methods for bayesian filtering", ACM Journals on Statistics and Computing, vol. 10, no. 3, pp.197-208, July 2000.
F. Gustafsson, F. Gunnarsson, N. Bergman, U. Forssell, J. Jansson, R. Karlsson, P. Nordlund, "Particle filters for positioning, navigation, and tracking", IEEE Transactions on Signal Processing, vol. 50, no. 2 pp. 425-437, Feb. 2002.
F. Evennou, F. Marx, "Advanced integration of WiFi and inertial navigation systems for indoor mobile positioning", EURASIP Journals on Applied Signal Processing, vol. 2006, pp.164-174, Jan. 2006.
R. Jirawimut, P. Ptasinski, V. Garaj, F. Cecelja, W. Balachandran, "A method for dead reckoning parameter correction in pedestrian navigation system", IEEE Transactions on Instrumentation and Measurement, vol. 52, no. 1, pp.209-2015, Feb. 2003.
L. Fang, P. Antsaklis, L. Montestruque, B. McMickell, M. Lemmon, Y. Sun, H. Fang, I. Koutroulis, M. Haenggi, M. Xie, X. Xie, "Design of a wireless assisted pedestrian dead reckoning system - the NavMote experience", IEEE Transactions on Instrumentation and Measurement, vol. 54, no. 6, pp.2342-2358, Dec. 2005.
A. Davison, I. Reid, N. Molton, O. Stasse, "MonoSLAM: Real-time single samera SLAM", IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 29, no. 6, pp.1052-1067, Jun. 2007.
C. Tsai, "A localization system of a mobile robot by fusing dead-reckoning and ultrasonic measurements", IEEE Transactions on Instrumentation and Measurement, vol. 47, no. 5, pp.1399-1404, Oct. 1998.
J. Hightower, G. Borriello, "Particle filters for location estimation in ubiquitous computing: A Case Study", in Proc. ACM Int. Conf. Ubiquitous Computing 2004 (ACM UbiComp 2004), pp.88-106, Nottingham, UK, Sep. 2004.
A. Paul, E. Wan, "RSSI-based indoor localization and tracking using sigma-point Kalman smoothers", IEEE Journal of Selected Topics in Signal Processing, Vol. 3, no. 5, pp.860-873, Oct. 2009.
J. Yim, S. Jeong, J. Joo, C. Park, "Utilizing map information for WLAN-based Kalman filter indoor tracking", in Porc. IEEE Int. Conf. Futrue Generation Communications and networking Symposia (IEEE FGCNS 2008), pp.58-62. Hainan Island, China, Dec. 2008.
A. Bekkali, H. Sanson, M. Matsumoto, "RFID indoor positioning based on probabilistic RFID map and Kalman filtering", in Proc. IEEE Int. Conf. Wireless and Mobile Computing, Networking and Communications 2007 (IEEE WiMOB 2007), pp.21-27, White plains, USA, Oct. 2007.
H. Lemelson, M. Kjaergaard, R. Hansen, T. King, "Error estimation for indoor 802.11 location fingerprinting", in Proc. ACM Int. Symp. Location and Context Awareness 2009 (ACM LoCA 2009), pp.138-155, Tokyo, Japan, May 2009.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.