Multi-Dimensional Wi-Fi Received Signal Strength Indicator Data Augmentation Based on Multi-Output Gaussian Process for Large-Scale Indoor Localization

Abstract

Location fingerprinting using Received Signal Strength Indicators (RSSIs) has become a popular technique for indoor localization due to its use of existing Wi-Fi infrastructure and Wi-Fi-enabled devices. Artificial intelligence/machine learning techniques such as Deep Neural Networks (DNNs) have been adopted to make location fingerprinting more accurate and reliable for large-scale indoor localization applications. However, the success of DNNs for indoor localization depends on the availability of a large amount of pre-processed and labeled data for training, the collection of which could be time-consuming in large-scale indoor environments and even challenging during a pandemic situation like COVID-19. To address these issues in data collection, we investigate multi-dimensional RSSI data augmentation based on the Multi-Output Gaussian Process (MOGP), which, unlike the Single-Output Gaussian Process (SOGP), can exploit the correlation among the RSSIs from multiple access points in a single floor, neighboring floors, or a single building by collectively processing them. The feasibility of MOGP-based multi-dimensional RSSI data augmentation is demonstrated through experiments using the hierarchical indoor localization model based on a Recurrent Neural Network (RNN)—i.e., one of the state-of-the-art multi-building and multi-floor localization models—and the publicly available UJIIndoorLoc multi-building and multi-floor indoor localization database. The RNN model trained with the UJIIndoorLoc database augmented with the augmentation mode of “by a single building”, where an MOGP model is fitted based on the entire RSSI data of a building, outperforms the other two augmentation modes and results in the three-dimensional localization error of 8.42 m.

On the Use and Construction of Wi-Fi Fingerprint Databases for Large-Scale Multi-Building and Multi-Floor Indoor Localization: A Case Study of the UJIIndoorLoc Database

“…Such clear coverage gaps between the datasets can negatively affect the model’s localization performance on the validation dataset. Data-augmentation techniques can be used in this regard to fill the spatial gaps in the training dataset and thereby improve the generalization capabilities and localization performance of a model [ 25 ].…”

“…Note that the position of a user’s phone (e.g., height and direction) affects RSSI measurements as does the model and software/firmware version of the phone; at the exact same location, a combination of different user and phone could result in different RSSI values. Given the dominance of a few users and phones in constructing the database, we can also apply techniques like data augmentation [ 25 ] and data resampling [ 26 ] to reduce the risk of bias and overfitting and thereby improve a model’s generalization as discussed in Section 3.1.1 .…”

“…To handle data imbalance, we can apply data-augmentation and/or data-resampling techniques to existing fingerprint databases: as for data augmentation, straightforward application of conventional techniques (e.g., [ 25 , 38 ]) could provide satisfactory results when the original records already have a good space coverage like those of Building 0 shown in Figure 5 b. When a building structure is complicated and original records poorly cover the space like the top floor of Building 2 or the bottom two floors of Building 1 of the UJIIndoorLoc database, however, we need a more sophisticated data-augmentation schemes like generative adversarial network (GAN)-based ones [ 39 ].…”

“…As discussed in Section 3.1.1, the UJIIndoorLoc database shows the uneven numbers of records per building and floor due to the accessibility of the restricted spaces like chemical laboratories and private offices [18]. We can apply post-construction techniques as practical alternatives in this regard: for example, we can utilize spatial data augmentation [25] for Building 0 and 1 and stratified sampling [49] for Building 2. As a smaller number of RPs with balanced space coverage is better than a larger number of RPs with poor space coverage even for the application of the post-construction techniques, the selection of RPs was and remains a top priority for the construction of fingerprint databases.…”