An indoor multi-source fusion positioning approach based on PDR/MM/WiFi

Abstract

The use of smartphones for indoor positioning has become increasingly popular in recent years. The cumulative error is an unavoidable problem for pedestrian dead reckoning (PDR). Mismatching is the main problem for fingerprint matching. Therefore, how to integrate multiple sensors to reduce the position error and improve the robustness of the positioning system is a subject worthy of further research. For fingerprint matching, an enhanced dynamic time warping (EDTW) is proposed to improve the accuracy of magnetic field matching. For PDR/magnetic matching (MM)/WiFi, a multi-source fusion positioning approach based on a robust extended Kalman filter (REKF) that introduces the estimation of the innovation sequence covariance is studied. The PDR-based error model is taken as the state transition equation and the difference between PDR and MM and the difference between PDR and WiFi as the observation equation. The experimental results show that the multi-source fusion positioning approach not only reduces the position error but also improves the robustness.

Introduction

In the outdoor open environment, global satellite navigation (GPS, Beidou, Glonass, Galileo) has been widely used [1], [2], [3]. The basic principle of satellite positioning is that the satellite transmits electromagnetic signals to the receiver. The receiver receives more than four satellite signals to calculate the specific position in the earth coordinate system. Although satellite navigation has been very successful outdoors, it is difficult to use satellite signals to locate indoors. The main reason is that it is difficult for low-power satellite signals to penetrate solid obstacles such as buildings, which causes indoor terminal equipment not to receive satellite signals most of the time. Therefore, it is very necessary to study other indoor positioning schemes.

With the acceleration of urbanization and the increasing number of large buildings, people spend 80% to 90% time indoors. To meet the needs of indoor localization, indoor positioning research has been concerned by the majority of researchers. Due to the variety of indoor positioning scenes, there are many indoor positioning technologies. The classical positioning techniques include INS [4], WiFi [5], [6], RFID [7], UWB [8], [9], infrared [10], geomagnetism [11], map matching [12], and so on. Indoor positioning technologies based on infrastructure include RFID, Zigbee, Bluetooth, UWB. Infrastructure-free indoor positioning technologies include INS, WiFi, and magnetic fields. The positioning technology based on infrastructure needs to set up signal transmitting equipment according to the indoor space environment. Different indoor spatial structures and environments require different numbers of signal transmitting equipment and different layouts. Considering the difference in the environment of each indoor place, there is not only the problem of high cost but also the problem of unified planning, which limits the popularity of infrastructure-based indoor positioning technology. The infrastructure-free technology mainly uses the sensor itself to obtain the motion information of the carrier or locates based on the existing infrastructure. One of the advantages of infrastructure-free positioning technology is that it does not require additional equipment, which determines that the positioning cost is relatively low and does not increase the additional costs of developers or consumers.

Accelerometer and gyroscope sensors have been embedded in smartphones. The spatial position of the user can be estimated by double integral acceleration and integral angular velocity. However, for cheap MEMS acceleration and gyroscope sensors, the integral operation easily increases the position error [13]. To reduce the error caused by direct integral acceleration and angular velocity, Gait-based PDR is often used to estimate the position of users in two-dimensional space [14]. Gait-based PDR includes three parts: step length, direction estimation, and step counting. Step length of pedestrians is affected by many factors, such as terrain, environment, height, weight, sex, age, etc. According to the collected acceleration, the step length model is divided into the parameter model and parameter-free model. The parameter model (e.g., the Weiberg model [15]) needs to train the model parameters to adapt the parameters to specific pedestrians. The parameter-free model estimates the user’s step length according to the collected acceleration, which does not need training parameters. The direction of the user is estimated by the integral angular velocity, but the direction error is also enlarged. To reduce the direction error, Wonho Kang proposed smartPDR, which combines the magnetic field and angular velocity to estimate the pedestrian direction [15]. A person’s walking posture is approximately periodic. Step counting of the user is estimated by dividing the vertical acceleration. The commonly used algorithms include peak detection [16], zero-crossing [17], autocorrelation [18], dynamic time warping (DTW) [19]. PDR estimates the current position based on historical location, step length, and direction. However, the cumulative error is an unavoidable problem. Many papers combine floor plans to eliminate the cumulative error [20].

The use of the magnetic field for positioning is another indoor positioning technology [21]. In the offline phase, the magnetic field fingerprint database is constructed in advance. In the online phase, the magnetic field collected by the smartphone searches similar magnetic fingerprints in the fingerprint database, the corresponding position of which is usually used as the location of the user. Mismatching is the main problem of magnetic field fingerprint matching. The changes in positioning environment and attitude angle of smartphones cause the fluctuation of the magnetic field. Therefore, magnetic field mismatching is inevitable.

Because many places (e.g., classrooms, shopping malls, libraries, office buildings) have installed many access points (APs), using WiFi for indoor positioning is becoming more and more popular [22]. The signal propagation model [23] and the fingerprint matching model [24] are the two main technical schemes at present. The signal propagation model needs to set the propagation parameters in advance [25]. Because the indoor environment changes with time, the propagation model inevitably brings positioning errors. WiFi fingerprint matching model is to associate the spatial position of the positioning region with the WiFi fingerprint; one location corresponds to a unique WiFi fingerprint. In the online positioning phase, it is necessary to use the collected signal features to search for a similar fingerprint in the fingerprint database. The corresponding position of the similar fingerprint is used as the estimated position. Similar to magnetic field fingerprint matching, mismatching is the main problem of WiFi fingerprint matching. At present, accelerometers, gyroscopes, magnetometers, and WiFi chips embedded in smartphones are low-cost microelectromechanical products. There are unavoidable problems in using a signal source for positioning. A reasonable scheme is that the positioning system integrates multiple source information to improve positioning accuracy. The REKF-based multi-source fusion positioning approach is proposed to fuse multiple sensor data for localization. The main innovations of the work are as follows.

(1) Considering that the direction of the smartphone is not the same in the offline phase and the online phase and the change of the environment, using the magnetic field amplitude as a one-dimensional magnetic fingerprint has stability. However, the spatial resolution of the magnetic fingerprint is reduced. Using the vertical and horizontal components of the magnetic field as the two-dimensional magnetic fingerprint can improve the spatial resolution of the fingerprint, but the stability of the fingerprint depends on the inertial sensor. Therefore, an EDTW algorithm is proposed in this paper, which combines the vertical and horizontal components of the magnetic field and magnetic field amplitude.

(2) Aiming at the outliers in the observation of multi-source fusion navigation system, REKF based on innovation sequence is used to reduce the large position error.

(3) Two pedestrians used two smartphones to carry out walking experiments in four experimental scenes. Experimental results show that the proposed algorithm can effectively improve the positioning accuracy and robustness.

The rest of the work is arranged as follows: Section 2 is related to the related work. Section 3 describes the system model in detail. Section 4 discusses the experiments of different scenarios. Section 5 has carried on summary and future work.