Localization with Transfer Learning Based on Fine-Grained Subcarrier Information for Dynamic Indoor Environments

Abstract

Indoor localization provides robust solutions in many applications, and Wi-Fi-based methods are considered some of the most promising means for optimizing indoor fingerprinting localization accuracy. However, Wi-Fi signals are vulnerable to environmental variations, resulting in data across different times being subjected to different distributions. To solve this problem, this paper proposes an across-time indoor localization solution based on channel state information (CSI) fingerprinting via multi-domain representations and transfer component analysis (TCA). We represent the format of CSI readings in multiple domains, extending the characterization of fine-grained information. TCA, a domain adaptation method in transfer learning, is applied to shorten the distribution distances among several CSI readings, which overcomes various CSI distribution problems at different time periods. Finally, we present a modified Bayesian model averaging approach to integrate the multi-domain outcomes and give the estimated positions. We conducted test-bed experiments in three scenarios on both personal computer (PC) and smartphone platforms in which the source and target fingerprinting data were collected across different days. The experimental results showed that our method outperforms state-of-the-art methods in localization accuracy.

Physical Tampering Detection Using Single COTS Wi-Fi Endpoint

“…Evidently, all different RPs had different trends of peaks, and the trends were virtually retained even after days. The finding of our experiment resonated with [ 48 ], which evaluated the CSI amplitude at the same location with the same orientation.…”

“…Evidently, all different RPs had different trends of peaks, and the trends were virtually retained even after days. The finding of our experiment resonated with [48], which evaluated the CSI amplitude at the same location with the same orientation Next, RPi3B+ was turned 45° left and right from the previous measurement to Orientation 2 (OR 2) and Orientation 3 (OR 3). The training data were kept the same as the previous measurement while using the datasets of the new orientation as the testing data.…”

Indoor Safety Monitoring for Falls or Restricted Areas Using Wi-Fi Channel State Information and Deep Learning Methods in Mega Building Construction Projects

“…In recent years, as deep learning technology has become more and more mature, deep learning algorithms have gradually become one of the most promising methods to analyze applications such as CSI with a large amount of data [19][20][21][22][23][24]. Further, the current deep learning technology has proved that as long as the size of the CSI data set collected is adequate, even if the data set contains erroneous values or does not cover the full range of data, meaningful results will still be produced [22,23,[25][26][27]. In addition to the CSI-related hardware and software issues, the deployment time of such an indoor safety warning system, including the hardware installation time and the data collection and analysis time, cannot be long.…”

Sensing the care: advancing unobtrusive sensing solutions to support informal caregivers of older adults with cognitive impairment

“…This initializes a pre-trained model through the process of pre-training right over a large amount of CSI-specific source task data, which allows the pre-trained model to be more specialized in CSI properties, thereby better facilitating the process of fine-tuning the target task. For example, the work by Yin et al [257] employs TL on CSI data for temporal indoor localization, in which three features (CSI amplitudes, wavelet transformations, and shape correlation) characterizing fine-grained information in CSI data were extracted to minimize the distances between CSI readings using Transfer Component Analysis. At last, a modified Bayesian model was utilized to estimate persons' positions.…”