Mean Teacher based SSL Framework for Indoor Localization Using Wi-Fi RSSI Fingerprinting

Sihao Li,Zhe Tang,Kyeong Soo Kim,Jeremy S. Smith

2024-07-18

Abstract:Wi-Fi fingerprinting is widely applied for indoor localization due to the widespread availability of Wi-Fi devices. However, traditional methods are not ideal for multi-building and multi-floor environments due to the scalability issues. Therefore, more and more researchers have employed deep learning techniques to enable scalable indoor localization. This paper introduces a novel semi-supervised learning framework for neural networks based on wireless access point selection, noise injection, and Mean Teacher model, which leverages unlabeled fingerprints to enhance localization performance. The proposed framework can manage hybrid in/outsourcing and voluntarily contributed databases and continually expand the fingerprint database with newly submitted unlabeled fingerprints during service. The viability of the proposed framework was examined using two established deep-learning models with the UJIIndoorLoc database. The experimental results suggest that the proposed framework significantly improves localization performance compared to the supervised learning-based approach in terms of floor-level coordinate estimation using EvAAL metric. It shows enhancements up to 10.99% and 8.98% in the former scenario and 4.25% and 9.35% in the latter, respectively with additional studies highlight the importance of the essential components of the proposed framework.

Machine Learning

What problem does this paper attempt to address?

The problems that this paper attempts to solve are as follows: In multi - building and multi - floor environments, traditional Wi - Fi fingerprint positioning methods have scalability issues. In particular, when dealing with large - scale data, a large amount of labeled data is required for training, which results in a time - consuming and labor - intensive data collection and labeling process. Moreover, it is difficult for traditional methods to maintain and update the database to keep the data up - to - date, which is crucial for long - term indoor positioning services.

To solve these problems, the author proposes a new framework based on semi - supervised learning (SSL), using unlabeled fingerprint data to enhance positioning performance. Specifically, the framework solves the problems in the following ways:

1. **Utilizing unlabeled data**: Traditional supervised learning frameworks require a large amount of labeled data for training, while the new framework can utilize unlabeled data, thereby reducing the dependence on a large amount of labeled data and reducing the cost of data collection and labeling.
2. **Continuously expanding the fingerprint database**: The framework can continuously receive new unlabeled fingerprint data during the service process and incorporate it into the database, thereby achieving dynamic expansion of the database.
3. **Improving positioning accuracy**: By introducing the Mean Teacher model, the framework can significantly improve positioning accuracy in multi - building and multi - floor environments, especially in floor - level coordinate estimation.

### Main contributions
1. A new indoor positioning framework based on semi - supervised learning is proposed, which can better balance scalability, complexity, and device dependence.
2. This framework can handle databases with a mixture of in - house and outsourced data and volunteer contributions, and continuously expand the fingerprint database during the service process.
3. Experiments were carried out based on two well - known and structurally different indoor positioning models and the UJIIndoorLoc database to verify the effectiveness of the framework.
4. Module - level insights are provided to demonstrate the effectiveness and efficiency of the proposed framework.

### Experimental results
The experimental results show that, compared with the methods based on supervised learning, the proposed framework improves by 10.99% and 8.98% respectively in floor - level coordinate estimation, and also improves by 4.25% and 9.35% respectively in other scenarios. These results highlight the importance of each component in the framework.

In this way, this paper not only solves the scalability problem of traditional Wi - Fi fingerprint positioning methods in multi - building and multi - floor environments, but also provides a more efficient and accurate indoor positioning solution.