US-12627948-B2 - Device-free localization methods within smart indoor environments

Abstract

Device-free localization for smart indoor environments within an indoor area covered by wireless networks is detected using active off-the-shelf-devices would be beneficial in a wide range of applications. By exploiting existing wireless communication signals and machine learning techniques in order to automatically detect entrance into the area, and track the location of a moving subject within the sensing area a low cost robust long-term tracking system can be established. A machine learning component is established to minimize the need for user annotation and overcome temporal instabilities via a semi-supervised framework. After establishing a robust base learner mapping wireless signals to different physical locations from a small amount of labeled data; during its lifetime, the learner automatically re-trains when the uncertainty level rises significantly. Additionally, an automatic change-point detection process is employed setting a query for updating the outdated model and the decision boundaries.

Inventors

• NEGAR GHOURCHIAN
• Michel ALLEGUE MARTINEZ
• Doina Precup

Assignees

• AERIAL TECHNOLOGIES INC.

Dates

Publication Date

20260512

Application Date

20231005

Claims (4)

1. 1 . A method for establishing target localization within an environment comprising: executing a first phase of a software model comprising an offline training phase; and executing a second phase of the software model comprising an online evaluation and adaptation phase; wherein the wireless signals are of a predetermined standard supporting communications between devices disposed with respect to the environment comprising at least a spatially separated transmitter and receiver where each wireless signal traverses a portion of the environment; the offline training phase configures the software model using a batch of labeled training data acquired from the environment comprising measured characteristics of wireless signals and location data relating to the location of a physical object within the environment at the time of measuring the characteristics of the wireless signals; and the online evaluation and adaptation phase comprises processing unlabeled streaming data measured characteristics of the wireless signals with the configured software model to establish physical location data of another physical object within the environment.
2. 2 . A method for establishing target localization within an environment comprising: executing a first phase of a software model comprising an offline training phase; and executing a second phase of the software model comprising an online evaluation and adaptation phase; wherein the wireless signals are of a predetermined standard supporting communications between devices disposed with respect to the environment comprising at least a spatially separated transmitter and receiver where each wireless signal traverses a portion of the environment; and the offline training phase comprises: receiving and analyzing wireless signals and their corresponding location labels whilst a user is present within different location spots of the environment; statistically formulating correlations between wireless signal readings and the location of the movements and events inside the environment through at least one algorithm of a plurality of algorithms, each algorithm relating to a step selected from the group comprising signal processing, data mining, and feature extraction; constructing a probabilistic localization model using a base classifier in dependence upon the statistically formulated correlations to generate respective decision boundaries and a confidence score that quantifies how certain the classifier is of its decision.
3. 3 . A method for establishing target localization within an environment comprising: executing a first phase of a software model comprising an offline training phase; and executing a second phase of the software model comprising an online evaluation and adaptation phase; automatically detecting at least one of a structural shift and a drift in the distribution of streaming data comprising: received live stream of unlabeled wireless signals arising from the communications between the devices disposed within the environment; applying a change-point-detection technique by continuously computing a divergence score; identifying the significant changes having a score above a predefined threshold; and outputting an indicator of drift to at least one of the first phase of the software model, the second phase of the software model, another system and another process; wherein the wireless signals are of a predetermined standard supporting communications between devices disposed with respect to the environment comprising at least a spatially separated transmitter and receiver where each wireless signal traverses a portion of the environment.
4. 4 . A method for establishing target localization within an environment comprising: executing a first phase of a software model comprising an offline training phase; and executing a second phase of the software model comprising an online evaluation and adaptation phase; wherein the wireless signals are of a predetermined standard supporting communications between devices disposed with respect to the environment comprising at least a spatially separated transmitter and receiver where each wireless signal traverses a portion of the environment; the first phase of the software model comprises establishing a base classifier for determining a location of a physical object within the environment in dependence upon wireless signals between devices disposed within the environment; and the second phase of the software model comprises adapting the decision boundaries of the base classifier with a process comprising the steps of: establishing at least one of a shift and a drift within the wireless signals and generating a drift indicator with the base classifier; triggering an active query system upon generation of the drift indication; generating a stream of confidence scores and predicted labels from the base classifier upon generation of the drift indicator; creating a repository of high-confidence samples of the wireless signals and their corresponding predicated labels in dependence upon the stream of confidence scores and predicted labels from the base classifier upon generation of the drift indicator; and updating the training data of the base classifier in dependence upon the high-confidence samples of the wireless signals and their corresponding predicated labels stored within the repository.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS This application claims the benefit of priority as a continuation of U.S. patent application Ser. No. 17/514,343 filed Oct. 29, 2021; which itself claims the benefit of priority as a continuation of U.S. patent application Ser. No. 17/019,759 filed Sep. 14, 2020 which has issued as U.S. Pat. No. 11,212,650; which itself claims the benefit of priority as a continuation of U.S. patent application Ser. No. 16/461,492 filed May 16, 2019 which has issued as U.S. Pat. No. 10,779,127; which itself claims priority as a 371 National Phase Application of PCT/CA2017/000,247 filed Nov. 21, 2017; which itself claims priority from U.S. Provisional Patent Application 62/425,267 filed Nov. 22, 2016; the contents of each being incorporated herein by reference. FIELD OF THE INVENTION This invention relates to localization and more particularly to systems, methods, and data processing apparatus for long-term and robust device-free localization in smart indoor spaces. BACKGROUND OF THE INVENTION Positioning or indoor localization is an essential function of a smart environment, which enables discovering valuable knowledge about the performances, behaviour and preferences of residents, especially those who need long-term monitoring or care. Moreover, location-based applications that utilize such information can offer customizable services according to the dynamics of their users' surroundings. Surveillance and security, health and sleep monitoring, assisted living for elderly people or patients with disabilities and entertainment are a few examples of applications wherein indoor location-aware computing has significantly improved performance. Generally, there are two different categories of indoor localization systems based on how their sensing infrastructure interacts with the target: device-based and device-free. Most approaches within the prior art exploit device-based systems, where the location of a moving target or human body within the space is determined and represented by a device associated with the moving target or human user such as a Wireless enabled smart phone or a radio-frequency identification (RFID) tag. These technologies are usually accurate and reliable, but most of them suffer from practical issues such as privacy concerns, physical contact with sensors, high implementation and maintenance cost, and cooperation from the subjects. Conversely, device-free passive (DFP) approaches do not require users to carry any devices or actively participate in the positioning process. Most of the DFP localization systems adopt a radio frequency (RF) sensing infrastructure (such as RFID, microwave, FM signals, etc.) and rely on the influence of the human body's presence and movement to influence these signals, e.g. through reflection. A few existing systems have employed information gleaned from Wireless signals such as channel state information (CSI) and received signal strength indicator (RSSI) to perform active or passive localization indoors. These systems are mainly enabled by recent wireless technology improvements and the fact that wireless signals are pervasive at most of indoor spaces such as residential, industrial, and public places. The basic idea amongst such systems is to take advantage of these wireless signals to monitor and quantify the distortions arising in the strength and patterns of signals between two nodes of communication (transmitter and receiver) and characterize the environment including human movements and their locations. See, for example, Xiao et al. in “Passive Device-Free Indoor Localization using Channel State Information” (Proc. IEEE 33rd Int. Conf. Distributed Computing Systems (ICDCS), pp. 236-245, 2013) wherein a CSI-based localization system utilizes multiple pairs of transmitter-receiver devices to estimate the location of a moving entity within a sensing area. Despite some preliminary success, most of these device-free passive systems have been implemented and evaluated using several devices in controlled sensing environments, such as a university laboratory or a classroom, with a large volume of human annotated data and within predefined and short-term scenarios. On the other hand, wireless signal components are sensitive to many internal and external factors including but not limited to multi-path interference, building attenuation, device and/or antenna orientation issues, changes in the environment (such as changing the position of objects) and signal interference. Therefore, performance of such localization systems usually degrades under realistic conditions and/or over time. Accordingly, it would be beneficial to provide a system that offers a robust and passive solution for inferring the location of a moving target within an indoor sensing area, which can be created by (at least) a pair of off-the-shelf wireless devices. Furthermore, it would be beneficial for the system to exploit a semi-supervised learning framework employing multipl