Combining passive Wi-Fi sensing and machine learning systems to monitor health, activity and well-being in nursing homes

Author : Dr Bo Tan, Electrical/Electronic Engineering Lecturer at Coventry University

12 February 2018

Figure 1: Passive Wi-Fi sensing for in-home activity sensing

Coventry University researchers explain how they used National Instruments LabVIEW and USRP to create a passive Wi-Fi sensing system that can detect one's body movements and vital signs through walls – and without any physical contact.

For the digital issue of this article, please visit this link – or click here to register for EPDT's magazine.

Detailed analysis of the Wi-Fi signals that reflect off a patient reveals patterns, which can be served up to gesture recognition libraries and machine learning systems for classification of activities and to model lifestyle behaviour for healthcare applications. The primary aim of health-related artificial intelligence (AI) applications is to analyse the relationships between prevention or treatment techniques and patient outcomes.

Such systems need to work with accurate signal data about instant and long-term activities that make up an individual’s pattern-of-life information. Engineers are currently exploring solutions for the following challenges in residential healthcare:

- Vital signs: Respiration and heart rate data can be accurately measured with a chest belt, electrocardiogram or photoplethysmography instruments. However, it is not practical to use such costly and inconvenient clinical instruments for daily monitoring scenarios in residential healthcare.

- Life-threatening events: Events like falling and slipping are chief causes of death in residential environments, especially for elderly or disabled people. Continuous surveillance can alert care workers when help is most needed.

- Daily activities: Monitoring a person’s daily activities can offer an abundance of health-related information. Even seemingly insignificant activities, like making a cup of tea, could provide information on water intake, sugar consumption and lifestyle. Sensing daily activities at home could unlock new insights for healthcare and human lifestyle research.

Chronically low-level activity awareness: Decreased physical activity can indicate signs of physical or mental health problems, such as chronic pain or depression. Therefore, weekly, monthly and long-term logs of activity levels can unlock a wealth of information for medical professionals – who can then determine root causes and preventative measures for health conditions.

The widely adopted detection methods currently used within care homes include wearable devices, camera-based vision systems and ambient sensors. However, these established options have major drawbacks (respectively physical discomfort, privacy concerns and limited detection accuracy).

There is an urgent requirement to develop novel monitoring solutions that are contactless, accurate and minimally invasive. This inspired our research into passive Wi-Fi sensing systems, and led us to adopt National Instruments (NI) LabVIEW and USRP (Universal Software Radio Peripheral) for rapid prototyping.

Passive Wi-Fi sensing technology 

The use of passive Wi-Fi sensing for residential healthcare is a natural extension of previous research at University College London (UCL), which proved the concept of passive Wi-Fi radar. Here, the term passive refers to the fact that users do not need to actively transmit a wireless signal to receive the radar echo. Instead, the passive Wi-Fi prototype, based on NI software-defined radio (SDR) solutions, leverages the wireless signals that already swamp our urban airways.

Because passive Wi-Fi radar is ‘receive only’, it is low power, unobtrusive and completely undetectable. This is a major benefit to military and counterterrorism scenarios. NI technologies best fit our research needs. The NI SDR solution was used to transition from concept to prototype to deployment faster than alternative approaches and is very versatile. We could repurpose the original prototype for entirely new applications, including health and activity monitoring in retirement and nursing homes.

However, scaling the prototype to fit the needs of residential healthcare required advancements in two key areas: signal processing and machine learning.

Through-wall respiration detection with USRP-based passive Wi-Fi sensing

Signal processing

Passive Wi-Fi sensing is a receive-only system that measures the dynamic Wi-Fi signal changes caused by moving indoor objects. The indoor multiple path propagation negatively impacts the wireless communication, but gives an opportunity for interpreting human activities. Due to the dynamic movement of a subject, the dynamic path presents time variation on the angle of arriving and propagation delay that correlate with the subject’s movements.

We can use frequency to measure phase changing rate during the measurement duration and Doppler shift to identify movements. We can discern real-time, high-resolution Doppler shifts for a given duration using batch processing boosted cross ambiguity function (CAF) analysis.

We also use the phase of each batch to identify small displacements of a subject, which can be used for inconspicuous body movement like breathing. Most commercial wireless network interface cards cannot deliver raw RF signal samples, which is why we choose USRP and LabVIEW software to capture, process and interpret the signals.

Machine learning

We can easily interpret some captured signals. For example, we can directly link the periodic change of batch phase with respiration rate. However, others may be difficult to understand visually. An example is the Doppler-time spectral map associated with gestures like picking things up or sitting down. Thus, we introduced machine learning to discover the link between the Doppler-time spectral map and physical activities. In practice, we tested principle component analysis (PCA) and singular value decomposition for Doppler-time spectral mapper feature extraction. Then we feed the features to support vector machine and sparse representation classifier (SRC).

The resulting classifiers show a promising capability of recognising the daily activities from Doppler-time spectral map. Besides the classification of the instant activities, the machine learning method can also model the pattern of a resident’s lifestyle by interpreting the long-term passive Wi-Fi activity data.

Building the proof-of-concept system

To prove the concept, we built a prototype system based on USRP SDR and LabVIEW. The USRP captures the raw IQ samples and delivers them to our LabVIEW application for rapid signal processing. LabVIEW delivers incredible flexibility for rapid prototyping, so we can adjust signal processing parameters to meet our exact requirements – whilst making the best use of multicore processing technologies. We could dynamically change the data arrays we work with, or alter the integration time and batch size of our analysis routines to adapt the system to slow and fast movements. Also, LabVIEW delivers an intuitive software environment, which empowered us to quickly integrate signal processing code (presented as subVIs in LabVIEW) for experimenting with new algorithms. In summary, NI offers an ideal platform to develop, test and verify cutting-edge pre-commercial concepts.

Experimental results

Based on the prototype system, we verified conceptual passive Wi-Fi sensing in two scenarios: activity recognition and through-wall respiration sensing.

Activity recognition

A group of gestures common in residential healthcare have been studied. Figure 3 shows the extracted Doppler-time spectral map from the two sensors during each gesture cycle. We then applied SRC classification based on the PCA features of each Doppler signature. Figure 3 shows the behaviour recognition result.

Using Wi-Fi radar to passively see through walls (see Supplementary information at bottom of this page)

Respiration sensing

As described in the previous section, the phase of the data batches is accurate enough to discern the small body movements caused by respiration. In this experiment, we demonstrated through-wall breathing detection. To observe the clear periodical signal varying caused by breathing, we used a Hampel filter to remove outliers and superfluous information.

Conclusion and impact

Compared to the established monitoring techniques (cameras, wearables and ambient sensors), passive Wi-Fi sensing displays obvious advantages:

- Contactless and pervasive: The ability to identify activities anywhere Wi-Fi connectivity is available, without the need for any subjects to carry devices.

- Diverse and accurate information: The detection of many activities from respiration to body gestures, from casual day-to-day operations to severe events.

- Unobtrusive: Because activity information is obtained from reflected RF signals and not images or video streams, we significantly reduce concerns over subject privacy.

Powering our research with USRP and LabVIEW has accelerated our groundbreaking research into passive Wi-Fi sensing – leading to innovative advancements, collaborations, patents, and more than 15 research publications. More importantly, the NI platform helped us realise the practical, commercial application of passive Wi-Fi sensing, and how this technology can positively impact people’s lives.

Supplementary information: using Wi-Fi radar to passively see through walls

The above project builds on previous work by Bo Tan and his team whilst at UCL, proving the concept that local Wi-Fi signals can be used to monitor moving objects and bodies that are otherwise visually obscured. Although fundamentally similar to traditional radar systems, relying on detecting the Doppler shifts in radio waves as they reflect off moving objects, this novel approach is entirely passive – utilising the wireless signals that already swamp our urban airways, rather than actively transmitting radio waves.

The complete lack of spectrum occupation and power emission ensures the radar is undetectable, making it ideal, for example, for efficient and undetectable military, security or public safety surveillance in urban settings.

Aside from public security applications (such as hijack or hostage situations), the passive detection could be applied in a broad range of scenarios, including crowd and traffic monitoring and human-machine interfacing.

Different types of wireless signals can be applied to different situations. For example, the system could acquire IEEE 802.11x (b, g, n, ac) signals to detect indoor moving targets for security purposes. Alternatively, the same system could monitor cellular signals, such as GSM or LTE, to detect direction and velocity of moving vehicles – before triggering an appropriate machine response to the detected movement.


Contact Details and Archive...

Print this page | E-mail this page