BERT-ADLOC: A secure crowdsourced indoor localization system based on BLE fingerprints
Introduction
Indoor positioning has been an active research front for almost two decades. Many signal modalities – radio signal in particular – have been extensively studied and fruitful results have been published. Certain pioneering business efforts [1] have been practiced, but large-scale deployment of indoor positioning systems are not yet to come.
The challenge of indoor positioning comes from the multi-path propagation of wireless signal and the complexity of indoor environments.
There are two trends in the research field. One trend is to develop more advanced analytical solutions [2], [3], [4], [5], [6], [7], [8], [9], [10]. They either seek to mitigate the multi-path effect by aggregating more bandwidths to better identify the direct path [5], [6], [7] or leverage the multi-path by forming multiple triangulations and resort to AoA principle for positioning [8]. Another trend is along the statistical way, where deep neural network (DNN) [11], convolutional neural network (CNN) [12], deep long short-term memory (DLSTM) [13] have been attempted in the past few years. Due to the superiority in extracting the deep hidden features, these deep learning models further improved the localization accuracy. Data required by these systems can be readily collected using commodity smartphones and are thus crowd-sensing friendly.
It is well known that building up a location database is extremely laborious and expensive. The two most labor-demanding efforts are data collection and labeling, which dominate the practicality of an IPS. In [14], the Gauss Process Regression (GPR) was applied to obtain the radio map of the entire space. As advocated in [15], [16], [17], [18], the most promising way is through crowdsourcing. The efficiency of crowdsourcing is evidently related to the number and also the involvement of participants. To this end, for efficient crowdsourcing, besides good incentive mechanism design, device requirement should be low, data labeling should be easy. Ideally, little or no labeling should be required.
Although many scholars have studied indoor localization systems based on crowdsourced environments, most of these researches only focus on indoor localization solution, ignoring the attacks that may be encountered in the realistic indoor positioning scene. As a large number of machine learning-based methods are used in indoor localization, the vulnerability of these approaches should not be ignored.
We have investigated recent researches on indoor positioning security. Wang et al. performed a comprehensive investigation of the security properties of ML algorithms under adversarial settings [19]. Tiku et al. [20] analyzed the possible attacks during the maintenance and update of the fingerprint database in a crowdsourcing environment and proposed the strategy for identifying malicious fingerprints. In [21], the malicious beacons attack during the online inference phase of the localization model was studied and a solution was designed. As far as we know, we are the first to design a complete system to resist attacks in the crowdsourcing scenario. We not only propose a reliable fingerprint database update scheme, but also provide a robust localization model that can resist online attacks in the crowdsourcing environment.
Our contributions are summarized as follows:
- •
We introduce tailored BERT as a powerful extractor to explore the underlying hidden deeper relation among BLE beacons through self-attention mechanism.
- •
We design BERT-AD to recognize malicious fingerprints in the fingerprint database update phase under crowdsourcing environment.
- •
We present robust indoor localization model BERT-LOCwhich can resist attacks in the online positioning phase and maintain good positioning results. These attacks include two aspects where valid beacons are moved and malicious beacons are deployed.
- •
For the first time, We present a secure crowdsourced indoor localization system BERT-ADLOC, which provides a reliable solution for updating the fingerprint database and ensures the robustness of the online location model in the presence of online attacks. We conduct multiple experiments to evaluate our proposed BERT-ADLOC.
The paper is organized in this order: Section 2 introduces the relevant work of indoor localization methods and indoor localization security. Section 3 analyzes multiple type of attacks during fingerprint database update and inference phase. Section 4 describes our proposed safe BERT-ADLOC localization system in detail. Section 5 shows experimental results and our discussions. Section 6 summarizes this paper.
Section snippets
Related work
This section briefly introduces the relevant work of BLE fingerprinting localization and indoor localization security. Because WiFi technology is earlier than BLE technology in the development of smart phones, BLE localization method directly draws on the Wi-Fi localization method. Although both Wi-Fi and BLE are in the 2.4 GHz band, the bandwidth of BLE signal is smaller than that of WiFi signal.
Many indoor localization technologies and applications have been proposed in the last two decades.
Adversary model and attacks
In this section, we first describe the adversary model in crowdsourcing environment and then analyze the potential attacks from adversaries in two aspects: fingerprints database update and malicious beacons in inference phase.
Methodology
In this section, we first describe an original feature extraction approach from BERT perspective, and then present an anti-attack positioning system BERT-ADLOC based on this idea. Our proposed BERT-ADLOC is applied in the crowdsourcing environment. There are two major cores in our system: the BERT-AD for adversarial samples recognizing and the BERT-based indoor localization model BERT-LOC. We first introduce the overview of the system, and the strategies regarding dealing with multiple
Experiments
In this chapter, we evaluate the proposed system BERT-ADLOC and compare it with other localization methods and attack identification methods in the crowdsourcing environment. We have designed and carried out experiments along the following two aspects:
- •
We conduct a set of experiments to test the BERT-AD’s correct recognition rate of the first four attacks we have discussed in the previous section, and compare it with other recognition strategies in [21] and [20]. In addition, we analyze the
Conclusion
In this paper, we focused on the vulnerability of the machine learning-based localization model. Aiming at defending against adversaries that may be encountered in the crowdsourced indoor localization scene, we analyzed the potential attacks from two main aspects: (1) during maintenance and updating of the fingerprint database (2) during online positioning. We initially proposed a complete and safe indoor localization system BERT-ADLOC in a crowdsourcing environment.
Our system includes the
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (46)
- et al.
The security of machine learning in an adversarial setting: A survey
J. Parallel Distrib. Comput.
(2019) - et al.
Bluetooth indoor positioning using rssi and least square estimation
IEEE ICFCC
(2010) - L. Li, G. Shen, C. Zhao, T. Moscibroda, J.-H. Lin, F. Zhao, Experiencing and handling the diversity in data density and...
- M. Kotaru, K. Joshi, D. Bharadia, S. Katti, Spotfi: Decimeter level localization using WiFi, in: Proceedings of the...
- et al.
Decimeter-level localization with a single wifi access point
- et al.
Widar: Decimeter-level passive tracking via velocity monitoring with commodity wi-fi
- et al.
Arraytrack: A fine-grained indoor location system
- et al.
Tonetrack: Leveraging frequency-agile radios for time-based indoor wireless localization
- et al.
Bloc: Csi-based accurate localization for ble tags
- et al.
Multipath triangulation: Decimeter-level wifi localization and orientation with a single unaided receiver
Locap: Autonomous millimeter accurate mapping of wifi infrastructure
Deep learning based wireless localization for indoor navigation
Applying deep neural network (dnn) for robust indoor localization in multi-building environment
Appl. Sci.
Cnn based indoor localization using rss time-series
Wifi fingerprinting indoor localization using local feature-based deep lstm
IEEE Syst. J.
Fast fingerprints construction via gpr of high spatial-temporal resolution with sparse rss sampling in indoor localization
Computing
Walkie-markie: Indoor pathway mapping made easy
Freeloc: Calibration-free crowdsourced indoor localization
Overcoming security vulnerabilities in deep learning–based indoor localization frameworks on mobile devices
ACM Trans. Embedded Comput. Syst. (TECS)
Secure crowdsourced indoor positioning systems
An improved indoor positioning algorithm based on rssi-trilateration technique for internet of things (iot)
Cited by (12)
Hybrid pedestrian positioning system using wearable inertial sensors and ultrasonic ranging
2024, Defence TechnologyExplainable indoor localization of BLE devices through RSSI using recursive continuous wavelet transformation and XGBoost classifier
2023, Future Generation Computer SystemsCitation Excerpt :The RSSI readings from multiple PSOC BLE nodes and the accompanying real 2-D locations of the supplied surveillance region are used to train the GRNN. Sun et al. discussed possible assaults during the updating of the fingerprint database and online inference phases and proposed the BERT-ADLOC indoor crowdsourced localization system, which is based on BLE fingerprints [31]. The adversarial sample discriminator BERT-AD and the indoor localization model BERT-LOC are the two primary components of the system.