An Improved WiFi Indoor Positioning Algorithm by Weighted Fusion

Rui Ma, Qiang Guo, Changzhen Hu, Jingfeng Xue

TL;DR

This paper tackles the challenge of low accuracy in WiFi-based indoor positioning systems, which are crucial for mobile applications. They propose a novel "weighted fusion" algorithm that combines an improved Euclidean distance metric and an improved joint probability method to achieve higher positioning accuracy than existing approaches.

What Problem Does This Paper Address?

Indoor positioning is a growing necessity, driven by the proliferation of mobile devices and the limitations of GPS indoors. While WiFi is a ubiquitous and low-cost technology for indoor positioning, its accuracy often falls short of practical application requirements, especially for scenarios demanding precision within 2 meters. Traditional WiFi indoor positioning methods, particularly those based on "location fingerprinting," face challenges with time-consuming data collection (offline stage) and insufficient accuracy (online stage). This paper aims to bridge this accuracy gap by enhancing both the fingerprint acquisition and the online positioning steps.

Methods

The proposed algorithm builds upon the traditional location fingerprinting approach and is divided into two main stages:

1. Offline Acquisition Process: * Signal Collection: WiFi signals are collected at a grid of equidistant points within the indoor environment. * Error Handling: Raw WiFi signals undergo error classification and handling to mitigate issues like systematic, gross, and random errors. This involves using offset values for hardware differences, T-tests for burst interference, and median filters for random fluctuations. A key detail is that missing signals are assigned a value of -100 dBm. * Database Construction: A database of location fingerprints is created, storing not just the average WiFi signal strength but also the standard deviation of the original WiFi signal strength and the average of the processed WiFi signal strength.

2. Online Positioning Process: * Pre-matching: To speed up positioning, a pre-match method is used to reduce the number of candidate fingerprints by selecting those with matching MAC addresses above a certain threshold. * Improved Euclidean Distance: This method is enhanced by incorporating the standard deviation of WiFi signal strength. This accounts for signal fluctuations, making it more robust than traditional Euclidean distance. It then selects the $K_d$ fingerprints with the smallest improved Euclidean distances to estimate an intermediate position $(X_1, Y_1)$. * Improved Joint Probability: This method is enhanced by using logarithmic calculations on the probability values. This helps to reduce the large differences in magnitude between probability values, making the calculation more sensitive and accurate. It then selects the $K_p$ fingerprints with the largest improved joint probabilities to estimate a second intermediate position $(X_2, Y_2)$. * Weighted Fusion: The final position $(X, Y)$ is calculated by combining the two intermediate results $(X_1, Y_1)$ and $(X_2, Y_2)$ using a weighted fusion method. The weights are determined by the variance of the intermediate results, giving more importance to the result that is more accurate (i.e., has a smaller variance).

Datasets

The paper describes experiments conducted in a real indoor environment (fourth floor of the software building at Beijing Institute of Technology). The data collected for this experiment constitutes the dataset used for evaluation. No specific public benchmark datasets are mentioned, but the experimental setup and data collection parameters are detailed.

Key Insights and Ideas

  • Signal Fluctuation Matters: The standard deviation of WiFi signal strength is a crucial, often overlooked, feature that directly reflects signal fluctuation. Incorporating this into distance calculations (like Euclidean distance) significantly improves robustness.
  • Logarithmic Transformation for Probabilities: Directly using raw probability values can be problematic due to their wide range. Applying a logarithmic transformation (as in the improved joint probability) makes the calculation more stable and sensitive to subtle differences.
  • Fusion for Robustness: Combining multiple positioning methods (Euclidean distance and joint probability) through weighted fusion leverages the strengths of each, leading to a more accurate and reliable final result than any single method alone. The weighting based on intermediate result variance is a clever way to dynamically adapt to the quality of each sub-method.
  • Error Handling is Key: Pre-processing raw WiFi signals to handle systematic, gross, and random errors is essential for building a high-quality fingerprint database and achieving better accuracy.

Key Results

The paper reports the following average positioning errors after 100 positioning trials:

  • Traditional WKNN algorithm: 1.66 m
  • WKNN with improved Euclidean distance: 1.60 m
  • Traditional joint probability algorithm: 1.93 m
  • Joint probability with improved joint probability: 1.87 m
  • Proposed Weighted Fusion Algorithm: 1.54 m

The results indicate that the proposed algorithm achieves the highest accuracy, outperforming the traditional WKNN and joint probability algorithms, as well as their improved versions. Figure 9 further illustrates that the proposed algorithm has a higher probability of achieving positioning errors within a given range (e.g., 90% of WKNN errors are < 2m, while the proposed algorithm achieves this with a higher probability).

Limitations and Future Work

Limitations: * Human Body Interference: The paper acknowledges that human movement and direction can affect WiFi signal strength (Figure 12). While they assume a "heading-oriented human direction is always to the front" to mitigate this, it's a simplification. * Environment Complexity: The paper notes that the difference in Euclidean distance between original and processed signals is small in their experiment due to a lack of people. They suggest that in more complex environments with more signal interference, the benefits of error handling would be more pronounced. * Multi-floor Environments: The current work uses a 2D model and does not explicitly consider multi-floor positioning.

Future Work: The authors highlight two key directions for future research:

  1. Automatic Fingerprint Database Construction: Developing methods to automatically build the fingerprint database, reducing the significant labor costs associated with manual site surveys.
  2. Database Maintenance without Human Interference: Once established, the database needs to be maintained and updated without requiring ongoing human intervention.

Summary updated

Abstract

The rapid development of mobile Internet has offered the opportunity for WiFi indoor positioning to come under the spotlight due to its low cost. However, nowadays the accuracy of WiFi indoor positioning cannot meet the demands of practical applications. To solve this problem, this paper proposes an improved WiFi indoor positioning algorithm by weighted fusion. The proposed algorithm is based on traditional location fingerprinting algorithms and consists of two stages: the offline acquisition and the online positioning. The offline acquisition process selects optimal parameters to complete the signal acquisition, and it forms a database of fingerprints by error classification and handling. To further improve the accuracy of positioning, the online positioning process first uses a pre-match method to select the candidate fingerprints to shorten the positioning time. After that, it uses the improved Euclidean distance and the improved joint probability to calculate two intermediate results, and further calculates the final result from these two intermediate results by weighted fusion. The improved Euclidean distance introduces the standard deviation of WiFi signal strength to smooth the WiFi signal fluctuation and the improved joint probability introduces the logarithmic calculation to reduce the difference between probability values. Comparing the proposed algorithm, the Euclidean distance based WKNN algorithm and the joint probability algorithm, the experimental results indicate that the proposed algorithm has higher positioning accuracy.
BibTeX
@inproceedings{ma2015improved,
  author = {Rui Ma and Qiang Guo and Changzhen Hu and Jingfeng Xue},
  title = {{An Improved WiFi Indoor Positioning Algorithm by Weighted Fusion}},
  booktitle = {Sensors},
  year = {2015},
  url = {https://openreview.net/forum?id=ptq4mndhb1},
}
Figure