1. Introduction
The development and popularization of communication technology have allowed the integration of indoor positioning technologies in smartphones [
1,
2,
3,
4]. The positioning services can be used for multiple purposes such as finding cars in underground parking lots [
5] and shopping in retail stores [
6]. However, owing to complex indoor environments, using a single technology to achieve positioning in various places is a challenging concept. Indoor positioning methods using multiple technologies such as Wi-Fi, Bluetooth, and audio have emerged [
7] to cater to varying requirements. In various technologies, smartphones can directly obtain Bluetooth low energy (BLE) signals using BLE beacons that can have separate batteries. The characteristics of low cost and simple deployment have popularized BLE beacons as a positioning solution for mobile phones [
8,
9,
10]. In addition, BLE beacon deployment rules are free, and different structures and densities can be deployed according to requirements. This enables range-based positioning technology to achieve good results [
10].
Indoor positioning and maps are usually used in conjunction to display positioning results to users. The user’s position in the scene is visually displayed by projecting the results onto the indoor map. Furthermore, integrating positioning technology with maps can provide improved indoor location services, such as check-in, indoor navigation, and other services [
2]. Map information can effectively constrain positioning results and paths, and is used in many areas of research to improve positioning results. In terms of positioning constraints, mapping elements such as doors and walls is critical to restrict and correct unreasonable results. Map information has been primarily applied to perform post-processing operations after acquiring positioning results. For example, using wall constraints to avoid positioning trajectories through the wall [
11,
12,
13], and utilizing basic heading information of buildings to reduce heading errors [
14,
15]. In terms of path constraints, it is critical to understand the relationship between topological connections and paths, and correct path deviations. To achieve this, previous studies have estimated the user’s route from the location estimation results, then considered the building’s passable areas to construct a link node topology map to correct the estimated route [
16,
17]. However, this method contains two processes of location estimation and trajectory estimation, which can easily lead to errors [
18]. Yamamoto et al. proposed that the received signal strength indicator (RSSI) of BLE can be combined with map information to directly estimate the global optimal path [
18]. This method does not require position estimation, and effectively improves the accuracy.
Different from the constraints of the map on the positioning results and trajectory in the existing research, we found that map information also has a constraint on the beacons. When the beacons are displayed on a map, the distribution and connectivity of the beacons are clear at a glance. For example, we can directly see which areas are more densely distributed and where the distribution is more reasonable. If this information can be modeled as prior knowledge, the beacon selection and result evaluation during positioning can be optimized. Thus, in practice, the usability of the positioning algorithm can be enhanced. On the one hand, the best combination of reference beacons can be selected in different areas. On the other hand, different beacon deployment structures and densities are given different evaluations. For example, densely deployed areas usually get better positioning results than sparse areas and so are evaluated as better.
In an ideal experimental environment, when beacons were deployed in a regular structure and reasonable spacing, optimal results were usually achieved [
19]. However, in real-world conditions, the physical environment is usually more complicated, which results in beacons not being deployed according to target rules and the disruption of signals. For these reasons, the signal composition for real-time positioning is extremely complicated.
Figure 1 shows the actual deployment of BLE beacons in an underground parking lot. The deployment of beacons and the signal coverage in each region were different. Theoretically, the density and structure of Area-D are optimal, and the best positioning accuracy can be obtained. However, the deployments Area-A and -B are not suitable for the trilateration positioning algorithm. Modeling the algorithm to select the appropriate beacon and algorithm for positioning according to the beacons around the device greatly affects the positioning effect in practical applications.
Owing to the complex structure of beacon deployment, it is difficult to determine a suitable positioning method and evaluate the availability of beacons from only real-time beacon sequences. This paper proposes a positioning method for BLE beacons aided by map information and improves the positioning effect by selecting appropriate reference beacons and positioning algorithms. The visual relationship between adjacent beacons in the area is modeled through a combination of map information and BLE location; a large positioning area is divided into multiple small cell areas with their respective layout shapes (points, lines, and polygons). Each cell area has a set of optimal reference beacons whose combinations are constructed using spatial relationships. A BLE beacon signal coverage rating map is constructed and divided into high, medium, and low levels to quantify the BLE beacon coverage of each cell area. In the positioning stage, sequences with severe signal fluctuations are eliminated according to the degree of suitability of the received signal strength indicator (RSSI) ranking of real-time beacons and their distribution in the actual space. Next, a reference beacon is selected from the relatively stable beacon sequences for positioning. Experiments demonstrate that this method can effectively improve positioning in a complex beacon deployment environment. Thus, a complete positioning algorithm framework is developed. The algorithm can determine the cell area based on the real-time beacon sequence and can accurately select the appropriate algorithm and reference beacon to improve positioning.
Section 2 introduces related research on BLE positioning and map-added positioning.
Section 3 describes the composition of the algorithm in detail.
Section 4 describes the experimental setup and the results.
Section 5 discusses the algorithm and experimental performance of this study, and finally summarizes the work of this paper and provides suggestions for future research.
2. Related Research
Current approaches to indoor localization can be categorized as infrastructure-based and infrastructure-less [
20]. Infrastructure-based approaches usually require existing infrastructures such as WIFI and BLE, and the use of smartphones to receive the RSSI of the beacon to achieve positioning [
21]. The RSSI reflects the distance between the device and the beacon. However, since RSSI is easily affected by environmental obstacles and multipath factors [
21,
22,
23], a filtering algorithm is usually required [
24,
25,
26] to enhance the accuracy of RSSI. In addition, the deployment structure of the beacons also greatly affects the availability of RSSI [
27]. Increasing the deployment density of beacons can ensure that sufficient reference beacons are within the visual range of the device [
28]. Moreover, high density will cause significant cross-talk between the beacons [
29].
Due to the complexity of the actual signal environment, the optimal selection of reference beacons or positioning algorithms is useful for accurate positioning [
30,
31,
32,
33]. The approximate point-in-triangulation (APIT) algorithm [
32] starts from the position of the beacon, and decides whether to filter the reference beacon by judging whether the unknown node is located in a triangle formed by any three beacons, the accuracy is greatly improved compared to before the optimization. Orujov et al. compared several positioning algorithms, then proposed and implemented a fuzzy logic based scheme to select the most fitting algorithm depending upon the size of the room, the number of beacons available and the strength of the signal [
33]. In follow-up research, they further used fuzzy logic type-2 to optimize the fingerprint positioning algorithm [
20]. This method obtained an average positioning accuracy of 0.43 m, which also verified the high applicability of the fingerprint positioning algorithm. Existing studies have shown that selecting appropriate positioning algorithms and reference beacons can effectively improve positioning accuracy. However, there is still a lack of effective application of prior information such as the spatial distribution relationship of beacons. In this study, we modeled the spatial relationship of beacons in a scene, and explored how this type of information can be used for the optimization of beacons and positioning algorithms.
Infrastructure-less approaches require no support from the existing infrastructure. For example, some studies use the built-in sensor of mobile phones to achieve the continuous positioning of pedestrians [
34,
35] or use geomagnetism for fingerprint positioning [
3]. In the realization of continuous positioning, algorithms such as Kalman filter [
12] and particle filter [
11,
13] are usually used to achieve the fusion of different positioning methods. Indoor maps are used to constrain the emergence of problems such as wall penetration during the fusion process [
11,
12,
13,
36]. Wang et al. proposed a maximum likelihood particle filter algorithm, which merges the particle prediction and updates steps into a single step, and avoids the waste of unnecessary particles as in the classic particle filter algorithm [
37]. However, the constraint effect of the map information on the reference beacon is rarely studied. By using the relationship between the scene and the beacon, the rationality of the combination of beacons can be restricted. For example, in a room, the beacon combination with the strongest RSSI is usually the beacon deployed inside the room. Here, we investigated the constraints the map information has on the combination of reference beacons, and how the independent reference beacons are used in beacon combinations.
5. Discussion
The map information-assisted BLE positioning algorithm proposed in this paper verifies that by selecting good BLE sequences for positioning and discarding poor ones, the accuracy of positioning can be effectively guaranteed. The real-time BLE sequence is relatively disrupted when the RSSI is affected by the environment, but when the BLE sequence is highly available, the RSSI is consistent with the actual distance.
By dividing the positioning area into multiple cell areas for analysis, the positioning algorithm and optimal reference beacon can be determined by area, which is more suitable for actual needs. For example, different areas require positioning with different accuracies, and beacons are deployed in different situations, such as single-point or multi-point deployment. In addition, in a multi-source scene where there are other positioning sources such as audio and light, the use of map information to assist area division can assist collaborative management of different sources in the multi-source scene, while enabling different positioning technologies in different areas.
However, there are still some limitations in this study. First of all, cell area recognition and reference beacon selection rely on the real-time signal sequence. If the signal contains large noise jitter, this algorithm will easily fail. Second, this study improves positioning by optimizing reference beacons without directly optimizing RSSI and positioning solution algorithms. In a real-world application, it must be used in conjunction with related algorithms. Finally, our algorithm is based on distance-based positioning. As fingerprint algorithms usually have better accuracy, the applicability of the algorithm to fingerprint positioning requires further research.