Large enterprises are often interested in tracking objects and people within buildings to improve resource allocation and occupant experience. Infrastructure-based indoor positioning systems (IIPS) can provide this service at low-cost by leveraging already deployed Wi-Fi infrastructure. Typically, IIPS perform localization and tracking of devices by measuring only Wi-Fi signals at wireless access points and do not rely on inertial sensor data at mobile devices (e.g., smartphones), which would require explicit user consent and sensing capabilities of the devices.
Despite these advantages, building an economically viable cost-effective IIPS that can accurately and simultaneously track many devices over very large buildings is difficult due to three main challenges. First, Wi-Fi signal measurements are extremely noisy due to unpredictable multipath propagation and signal attenuation. Second, as the IIPS obtain measurements in a best effort manner without requiring any applications installed on a tracked device, the measurements are temporally sparse and non-periodic, which makes it difficult to exploit historical measurements. Third, the cost-effective IIPS have limited computational resources, in turn limiting scalability in terms of the number of simultaneously tracked devices.
Prior approaches have narrowly focused on either improving the accuracy or reducing the complexity of localization algorithms. To compute the location at the current time step, they typically use only the latest explicit Wi-Fi measurements (e.g., signal strengths). The novelty of our approach lies in considering contexts of a device that can provide useful indications of the device's location. One such example of context is device motion. It indicates whether or not the device's location has changed. For a stationary device, the IIPS can either skip expensive device localization or aggregate noisy, temporally sparse location estimates to improve localization accuracy. Another example of context applicable to a moving device is a floor map that consists of pre-defined path segments that a user can take. The map can be leveraged to constrain noisy, temporally sparse location estimates on the paths.
The thesis of this dissertation is that embedding context-aware capabilities in the IIPS enhances its performance in tracking many devices simultaneously and accurately. Specifically, we develop motion detection and map matching to show the benefits of leveraging two critical contexts: device motion and floor map. Providing motion detection and map matching is non-trivial in the IIPS where we must rely only on data from the Wi-Fi infrastructure.
This thesis makes two contributions. First, we develop feature-based and deep learning-based motion detection models that exploit temporal patterns in Wi-Fi measurements across different access points to classify device motion in real time. Our extensive evaluations on datasets from real Wi-Fi deployments show that our motion detection models can detect device motion accurately. This, in turn, allows the IIPS to skip repeated location computation for stationary devices or improve the accuracy of localizing these devices. Second, we develop graph-based and image-based map matching models to exploit floor maps. The novelty of the graph-based approach lies in applying geometric and topological constraints to select which path segment to align the current location estimate. Our graph-based map matching can align a location estimate of a user device on the path taken by the user and close to the user's current location. The novelty of the image-based approach lies in representing for the first time, input data including location estimates and the floor map as 2D images. This novel representation enables the design, development, and application of encoder-decoder neural networks to exploit spatial relationships in input images to potentially improve location accuracy. In our evaluation, we show that the image-based approach can improve location accuracy with large simulated datasets, compared to the graph-based approach. Together, these contributions enable improvement of the IIPS in its ability to accurately and simultaneously track many devices over large buildings.
| Identifer | oai:union.ndltd.org:pdx.edu/oai:pdxscholar.library.pdx.edu:open_access_etds-6081 |
| Date | 04 June 2019 |
| Creators | Tran, Huy Phuong |
| Publisher | PDXScholar |
| Source Sets | Portland State University |
| Detected Language | English |
| Type | text |
| Format | application/pdf |
| Source | Dissertations and Theses |
Page generated in 0.0021 seconds