Finding Your Way with Wi-Fi
The FloorLocator method utilizes Wi-Fi signals and a spiking graph neural network for floor localization with a high level of accuracy.
Indoor positioning systems (IPS) are important for applications that require navigating indoor environments where GPS signals are unavailable or unreliable. They utilize a variety of technologies such as Wi-Fi, Bluetooth, RFID, ultrasonic, and inertial sensors to determine the location of objects or people within a confined space. The importance of IPS lies in its ability to provide accurate real-time location information, enabling a wide range of use cases including indoor navigation, asset tracking, location-based services, and context-aware computing.
In large, multistory buildings, an IPS must be supplemented by a floor localization system, which determines the specific floor number that an object occupies. However, existing floor localization technologies face significant challenges. Many rely on specialized sensors that are not commonly available in consumer devices, limiting their widespread adoption. Additionally, these methods often require extensive knowledge about the physical dimensions of the building and the precise positions of multiple transmitters, making deployment complex and costly. Furthermore, these techniques tend to be computationally intensive and lack scalability, rendering them impractical for many real-world use cases where efficiency and cost-effectiveness are crucial.
As a result, there is a growing demand for more robust and scalable floor localization solutions that can overcome these limitations and unlock the full potential of indoor positioning technologies. One recent entrant into the field was just unveiled by a team at Chongqing University. They have developed a method that leverages Wi-Fi signals from access points distributed throughout a building, and feeds that information into a machine learning algorithm to accurately pinpoint the floor number that an object is on. This method does not require any information about the layout of the building or the precise locations of the access points, and it is also computationally efficient and scalable.
The system, called FloorLocator, first scans for all nearby Wi-Fi signals. These signals are then arranged into a graph structure based on the proximity of the access point to the receiver. This arrangement of the data enables FloorLocator to operate even when the precise locations of access points are unknown. Next, the graph data is fed into a spiking graph neural network. This network architecture has the advantage of the computational efficiency of spiking neural networks, as well as the advanced pattern recognition capabilities of graph neural networks. This deep learning model is then tasked with predicting the most likely floor number that an object is located on.
To evaluate the performance of the system, the researchers ran a series of experiments in which FloorLocator was asked to predict the floor number that an object was located on. An average floor recognition accuracy of nearly 96 percent was observed. This is at least a 10 percent improvement over existing state of the art methods. It is not entirely clear how generalized the system is at this time, however, as the recognition accuracy dropped to 82 percent when working with data from a different building. But it is notable that this result still beats existing methods by about 4 percent.
As it presently stands, it is known that FloorLocator will be dramatically slowed down in very large buildings. The team intends to explore the possibility of constructing a more efficient input graph of access points to mitigate this problem. They also plan to implement their algorithm on neuromorphic hardware, which would be expected to further enhance the system’s computational efficiency.
R&D, creativity, and building the next big thing you never knew you wanted are my specialties.