TL;DR
This paper introduces a novel graph optimization approach to simultaneously estimate the positions of Bluetooth Low-Energy (BLE) beacons and create a reference fingerprint map (RFM) for indoor positioning. This method eliminates the need for costly and time-consuming manual surveying, making indoor positioning more accessible.
What Problem Does This Paper Address?
Indoor positioning is crucial for many applications, but existing methods using technologies like Bluetooth Low-Energy (BLE) beacons face a significant hurdle: they require pre-existing knowledge of beacon locations or a detailed reference fingerprint map (RFM) of the environment. Obtaining this information typically involves extensive manual surveying, which is labor-intensive, expensive, and prone to errors, especially in large indoor spaces. This paper tackles the challenge of efficiently and accurately estimating these essential pre-requisites without relying on dedicated surveying equipment.
Methods
The core of the proposed method is a graph optimization framework that leverages multiple data sources to estimate both user poses and beacon positions. Here's a breakdown:
- Sensor Fusion: The system integrates data from a BLE-enabled mobile device, specifically:
- Inertial Measurement Unit (IMU): Used for Pedestrian Dead Reckoning (PDR) to estimate relative changes in user pose (position and orientation) between consecutive steps. This generates "PDR constraints."
- BLE Received Signal Strength Indication (RSSI): Collected from nearby BLE beacons. These RSSI values are used in two ways:
- To estimate the distance to beacons via a path-loss model, creating "range-based constraints." * To identify similar "fingerprints" (sets of RSSI values from multiple beacons) at different locations, generating "fingerprint matching constraints."
- Graph Optimization: These constraints are formulated as terms in a cost function with a least-squares structure. The graph optimization framework then iteratively minimizes this cost function to find the optimal user trajectory and the precise locations of the BLE beacons. The RFM is implicitly generated as a byproduct of estimating user poses at various locations.
- Key Innovation: The novelty lies in the simultaneous estimation of both beacon positions and the RFM, eliminating the need for separate, manual surveying steps. The method uses readily available sensors on a smartphone.
Datasets
The paper describes experiments conducted in an office site of approximately 90m x 37m. They installed 48 BLE beacons and used ground truth positions measured by a total station for evaluation. For testing the positioning accuracy, they also set up 50 landmarks on the ground. The paper does not introduce a new benchmark dataset but uses a real-world environment for evaluation.
Key Insights and Ideas
- Eliminating Surveying Burden: The most significant insight is that accurate indoor positioning can be achieved without the laborious and costly process of manual beacon surveying. This dramatically lowers the barrier to entry for BLE-based indoor positioning systems.
- Synergy of Constraints: Combining PDR, range-based RSSI measurements, and fingerprint matching constraints proves more robust than using any single type of constraint alone. Each constraint type helps to mitigate the weaknesses of the others.
- RFM Generation as a Byproduct: The method elegantly generates the reference fingerprint map (RFM) as part of the pose estimation process, rather than requiring it as a separate, upfront task.
- Robustness to Inaccurate Pre-requisites: The experiments show that even with estimated beacon positions and an RFM, the positioning accuracy remains reasonably good, demonstrating the robustness of the graph optimization approach.
Key Results
- Beacon Position Estimation: The proposed method achieved mean positioning errors of approximately 1.27 m in a dense beacon scenario and 2.26 m in a sparse beacon scenario.
- Range-Based Positioning Accuracy: Using the estimated beacon positions, the mean positioning error was 3.25 m (dense) and 4.69 m (sparse). This is a slight degradation compared to using ground truth beacon positions (2.66 m and 4.03 m, respectively), but still considered acceptable.
- Fingerprinting-Based Positioning Accuracy: Using the estimated RFM, the mean positioning error was 2.78 m (dense) and 4.11 m (sparse). This performance was slightly better than the range-based method using estimated beacon positions.
- Comparison to Prior Work: While the proposed method's accuracy is slightly lower than some prior methods that rely on precise surveying (e.g., a mean error of 2.57m for range-based in [11]), it offers a significant advantage in terms of reduced effort and cost.
Limitations and Future Work
- Accuracy Trade-off: The paper acknowledges that the positioning accuracy is slightly lower than methods that use precisely surveyed beacon positions and RFMs. This is an inherent trade-off for the convenience of not requiring manual surveying.
- Impact of Walking Persons: The authors mention that the effect of walking persons in the environment is a factor that could be further investigated.
- Sophisticated Path-Loss Models: Exploring more advanced path-loss models could potentially improve range-based accuracy.
- Upper Bound Error: Further analysis of the upper bound error is suggested.
- Parameter Optimization: The paper notes that the thresholds for fingerprint-based constraints were adopted from previous work and suggests performing an exhaustive search for optimal parameters in the future.
- Number of Beacons: The impact of the number of beacons on performance is an area for future study.
Summary updated