Radar-based system and method for monitoring of subjects

Abstract

A system and method for monitoring a subject in a three dimensional environment comprising one or more rooms, the system and method comprising the use of one or more radar sensors for transmitting radar waves and for receiving reflections of the radar waves. The system and method comprise one or more processors for detecting and analyzing the reflected radar wave in order to monitor movements of the subject. The system and method can be used, for example, to monitor the presence, entry, and exit of the subject with respect to a space and to monitor the subject for falls. The system and method further comprise transmitting the detected and analyzed information to a user.

Description

RADAR-BASED SYSTEM AND METHOD FOR MONITORING OF SUBJECTSCROSS REFERENCE TO PRIOR APPLICATIONS

[0001] This application claims priority to US Application No. 63 / 571 ,412, filed March 28, 2024. The entire contents of such prior application are incorporated herein by reference.FIELD OF THE DESCRIPTION

[0002] The present disclosure relates generally to a system and method for monitoring location and movements of subjects, such as residents in facilities, such as hospitals, assisted living facilities, long-term care facilities, and the like. In one aspect, the present disclosure relates to a system and method for detecting and monitoring movements of one or more individuals and analyzing such data to predict and thereby prevent and / or treat falls and underlying medical conditions relating to individuals.BACKGROUND

[0003] One of the leading causes of serious injuries among seniors is falling. It is estimated that over 20 million undetected fall incidents occur among seniors in North America every year. It is estimated that 50% of seniors who suffer from undetected falls will die within 6 months of the fall. It is also estimated that a fall injury costs $30,000 on average. Early detection and response to a fall might reduce the risk of death by 88% and may also reduce long-term disability from a fall.

[0004] Falls are the primary preventable cause of death and serious injury among seniors. Current technologies are usually not viable for persons living with dementia and cannot be relied upon for predicting and thereby preventing falls. Even if the fall is detected, currently known systems may not adequately detect the severity of the fall, which leaves the decision to staff to judge the severity of the fall after the incident has occurred.

[0005] Long-term care, LTC, facilities face chronic staffing problems and are desperate for solutions that allow them to improve the quality of care without increasing headcount. Moreover, the high attrition rate amongst workers involved in frontline caregiver roles, especially in LTC facilities and retirement homes, results in chronic staff shortages and, consequently, reduced quality of care, with residents and staff being placed in even more precarious situations.

[0006] The imposition of metered “minimum daily average hours of direct personal care” is unsustainable. The amount of support required for individuals in care will continue to grow, and operators of facilities simply cannot afford the number of qualified staff this would require, nor are they available to do so. Even if the “minimum daily average hours of direct personal care” is somehow achieved, that still leaves at-risk individuals, especially seniors, unmonitored for the remaining hours of the day.

[0007] Technological monitoring solutions have been proposed in an effort to address the staffing shortage issues. Such solutions often involve the use of wearable sensor devices and visual observation devices involving optical and / or thermal cameras. Wearable devices function by detecting force, acceleration, or pressure applied to the sensor.However, these devices require frequent charging and are susceptible to damage. Such devices also pose an inconvenience for the individuals wearing same.

[0008] Although thermal cameras provide good imaging in low light intensity environments, they might cause interference with other medical devices. One of the best solutions is camera vision; however, camera-based systems raise privacy concerns and are sensitive to illumination levels and sunlight.

[0009] Radar sensors are appealing for healthcare applications, with potential benefits for older adults and healthcare systems. Radar-based systems offer a solution that is suitable for several applications because they preserve privacy, are not affected by illumination or sunlight, and, particularly if used in millimeter (mm) wave frequencies, do not interference with other medical devices or communication systems. Radar systems can be adjusted in terms of configurations and resolutions thereby offering advantages over other known systems. However, despite numerous studies on the application of radar sensors in healthcare, there is currently no system in place to adequately detect and / or predict fall incidents, analyze historical data, and so forth.

[0010] Some known radar-based systems are described in US 8,068,051 ; US 7,916,006; US 7,567,200; US 2013 / 0300573; and US 2021 / 0298643, the entire contents of which are incorporated herein by reference. While such known systems incorporate motion detection devices, such as radar systems, they primarily provide crude data with respect to object detection and tracking, which requires human intervention to confirm an event, such as the fall of a resident. These known systems therefore often result in false alarms that unnecessarily occupy limited staff time. Furthermore, many of these known systems do notallow for automatic interpretation of motion data to predict underlying conditions or that a fall is likely to occur.

[0011] There is a need for a solution to address at least one of the deficiencies in the known systems for monitoring individuals, such as residents or patients, in one or more locations.SUMMARY OF THE DESCRIPTION

[0012] In one aspect, there is provided a system for monitoring a subject in a three- dimensional environment comprising one or more rooms, the system comprising:

[0013] - one or more radar sensors, each of the radar sensors comprising at least one radar wave transmitter for transmitting radar waves and at least one radar wave receiver for receiving reflections of the transmitted radar waves;

[0014] - a processor configured to receive data from the one or more radar sensors, the data being indicative of the reflected radar signals received by the at least one radar wave receiver;

[0015] - a memory in communication with the processor, the memory having stored therein one or more executable algorithms for processing the data received from the one or more radar sensors;

[0016] - wherein the processor is configured to monitor the movement of the subject in the environment in three dimensions based on the data received from the one or more radar sensors.

[0017] In another aspect, the system is configured to detect the presence of a subject in a given space (e.g., a particular room), and the entry and exit of the subject from such space.

[0018] In another aspect, there is provided a method for monitoring a subject in a three- dimensional environment comprising one or more rooms, the method comprising:

[0019] - providing one or more radar sensors, each of the radar sensors comprising at least one radar wave transmitter for transmitting a radar signal and at least one radar wave receiver for receiving a reflection of the radar signal;

[0020] - transmitting the reflected radar signals to a processor;

[0021] - processing the reflected radar signals with a presence-absence detection algorithm to determine entry and exit activities of the subject from at least a first room of the environment.

[0022] In another aspect, there is provided a method for monitoring a subject in a three- dimensional environment comprising one or more rooms, the method comprising:

[0023] - providing one or more radar sensors, each of the radar sensors comprising at least one radar wave transmitter for transmitting a radar signal and at least one radar wave receiver for receiving a reflection of the radar signal;

[0024] - transmitting the reflected radar signals to a processor;

[0025] - processing the reflected radar signals with a fall severity detection algorithm to detect and analyze a fall incident of a subject.BRIEF DESCRIPTION OF THE FIGURES

[0026] Certain features of the present description will become more apparent in the following detailed description in which reference is made to the appended figures wherein:

[0027] Fig. 1 illustrates a typical radar sensor system, comprising generating electromagnetic waves and receiving the reflected signals.

[0028] Fig. 2 illustrates how radar coverage may be configured for a home.

[0029] Fig. 3 (a) and Fig.3 (b) demonstrate a method of determining a radar pattern and coverage inside a room.

[0030] Fig. 4 shows an example of two radar sensors installed on different sides of a wall in a room.

[0031] Fig. 5 (a), Fig. 5 (b), and Fig. 5 (c) illustrate a system comprising radar sensors in a bedroom, in a living room and in a washroom, respectively.

[0032] Fig. 6 illustrates an example of networking and data streaming of multiple sensors in a large facility.

[0033] Fig. 7 illustrates an example of a processing chain to generate Range Doppler (RDM), Range Azimuth (RAM), and Range Elevation (REM) heatmaps.

[0034] Fig. 8 shows a flowchart of a presence-absence detection algorithm.

[0035] Fig. 9 shows a flowchart of a get_position algorithm.

[0036] Fig. 10 illustrates an aspect of the sensor system.

[0037] Fig. 11a shows a flowchart of a suspected fall detection algorithm.

[0038] Fig. 11 b shows a further flowchart of the suspected fall detection algorithm.

[0039] Fig. 12 shows a flowchart of an activity recognition algorithm.

[0040] Fig. 13 illustrates an example of a sensor system for identifying adverse events using machine learning models.

[0041] Figs. 14(a) to 14(c) illustrate short-time Fourier transform (STFT) patterns of a subject (a) sitting on the sofa, (b) moving, and (c) walking.

[0042] Figs. 15(a) to 15(h) illustrate range-Doppler maps of (a) an empty room, (b) a subject sitting on a sofa, (c) a subject washing dishes, (d) a subject picking up an object, (e) a subject vacuuming, (f) a subject vacuuming, (g) a subject walking, and (h) a subject walking.

[0043] Fig. 16(a) shows range-Doppler maps of a person laying on a bed.

[0044] Fig. 16(b) shows a camera output of a person laying on a bed.

[0045] Fig. 16(c) range-Doppler maps of an empty room.

[0046] Fig. 16(d) shows a camera output of an empty room.

[0047] Fig. 17 shows an example of range-azimuth, range-elevation, and range-Doppler map of a fall incident.

[0048] Figs. 18(a) and 18(b) illustrate the performance of a trained model.

[0049] Fig. 19 shows a confusion matrix of a deep learning model.

[0050] Fig. 20 illustrates an example of walking tests for developing the described system.

[0051] Fig. 21 illustrates the range-time plot of a walking person.

[0052] Fig. 22 illustrates the maximum velocity of a walking person based on torso movement.

[0053] Fig. 23 illustrates step points, step length step time and number of steps of a walking person.

[0054] Fig. 24 illustrates how the sensor system collects information from different sources for formulating a decision.

[0055] Fig. 25(a) illustrates an example display screen on a handheld device showing Exception Reports showing abnormalities in washroom usage.

[0056] Fig. 25(b) illustrates an example display screen on a handheld device showing Exception Reports showing abnormalities in activity level.

[0057] Fig. 25(c) illustrates an example display screen on a handheld device showing Exception Reports showing abnormalities in movement.

[0058] Fig. 26(a) illustrates an example display screen on a handheld device showing Alerts and Confirmations: requesting to respond a suspected fall.

[0059] Fig. 26(b) illustrates an example display screen on a handheld device showing Alerts and Confirmations, showing no response for a suspected fall.

[0060] Fig. 26(c) illustrates an example display screen on a handheld device showing Alerts and Confirmations: requesting to confirm a suspected fall.

[0061] Fig. 27(a) illustrates an example display screen on a handheld device showing Alerts and Confirmations: requesting to respond a suspected bed movement.

[0062] Fig. 27(b) illustrates an example display screen on a handheld device showing Alerts and Confirmations: showing no response for a suspected bed movement.

[0063] Fig. 27(c) illustrates an example display screen on a handheld device showing Alerts and Confirmations: requesting to confirm a suspected bed movement.

[0064] Fig. 28(a) illustrates an example display screen on a handheld device showing a detection of severity of a fall, with 54% danger.

[0065] Fig. 28(b) illustrates an example display screen on a handheld device showing a detection of severity of a fall, with 12% danger.

[0066] Fig. 28(c) illustrates an example display screen on a handheld device showing a detection of severity of a fall, with 89% danger.

[0067] Fig. 29(a) illustrates a slow fall speed captured by the sensor.

[0068] Fig. 29(b) illustrates a fast fall speed captured by the sensor.

[0069] Fig. 30(a) illustrates an example dashboard on a display screen showing a home area view.

[0070] Fig. 30(b) illustrates another example dashboard on a display screen showing a home area view.

[0071] Fig. 31 (a) illustrates an example dashboard on a display screen showing a resident in-depth view.

[0072] Fig. 31 (b) illustrates another example dashboard on a display screen showing a resident in-depth view.

[0073] Fig. 32 illustrates an example of the designed dielectric lens antenna to increase the radar gain and sharpen the beam.

[0074] Fig. 33 illustrates a flow diagram of a fall detection prior to classification algorithm.

[0075] Fig. 34 is a front perspective view of a housing for a radar sensor according to an aspect of the description.

[0076] Fig. 35 is a rear perspective view of the housing of Fig. 34.DETAILED DESCRIPTION

[0077] In the present description, the terms “resident”, “individual”, “subject”, or “patient” may be used. It will be understood that these terms are used herein as equivalent terms andare not intended to limit the present description to any particular use. The system and method described herein may be used for monitoring the movements, status, or locations of any lifeform, such as a human or an animal. As will also be understood, the present description is particularly suited for use in monitoring patients and / or residents in a health or long-term care facility where their well-being is being handled.

[0078] Similarly, the present description may recite the terms “residence”, “facility”, “home”, “hospital”. It will be understood that such terms are used for convenience and are not intended to limit the present system and method to any specific location. The presently described system and method may be used for any location or purpose where monitoring of one or more subjects is desired. As will also be understood, the present description is particularly suited for use in patient or resident care is required.

[0079] The present description contains details of a system and a method. The terms “system” and “method” may be used alone or in combination. It will be understood that the description, whether referring only to a “system” or a “method”, applies equally to both aspects. Thus, the system comprises hardware components that may be utilized to perform the method, and vice versa.

[0080] The terms “comprise”, “comprises”, “comprised” or “comprising” may be used in the present description. As used herein (including the specification and / or the claims), and unless stated otherwise, these terms are to be interpreted as open-ended terms and as specifying the presence of the stated features, integers, steps or components, but not as precluding the presence of one or more other feature, integer, step, component or a group thereof as would be apparent to persons having ordinary skill in the relevant art. Thus, the term “comprising” as used in this specification means “consisting at least in part of’. When interpreting statements in this specification that include that term, the features, prefaced by that term in each statement, all need to be present, but other features can also be present. Related terms such as “comprise” and “comprised” are to be interpreted in the same manner.

[0081] The phrase “consisting essentially of’ or “consists essentially of’ will be understood as generally closed terms, with the exception of allowing inclusion of additional items, materials, components, steps, or elements, that do not materially affect the basic and novel characteristics or function of the item(s) used in connection therewith. For example, trace elements present in a composition, but not affecting the composition’s nature orcharacteristics would be permissible if present under the “consisting essentially of’ language, even though not expressly recited in a list of items following such terminology. When using an open-ended term, such as “comprising” or “including”, it will be understood that direct support should be afforded also to “consisting essentially of’ language as well as “consisting of’ language as if stated explicitly and vice versa. In essence, use of one of these terms in the specification provides support for all of the others.

[0082] For the purposes of the present description and / or claims, and unless otherwise indicated, all numbers expressing quantities, percentages or proportions, and other numerical values used in the specification and claims, are to be understood as being modified in all instances by the term “about.” Accordingly, unless indicated to the contrary, the numerical parameters set forth herein are approximations that may vary depending upon the desired properties sought to be obtained by the present invention, inclusive of the stated value and has the meaning including the degree of error associated with measurement of the particular quantity. The term “about” generally refers to a range of numbers that one of ordinary skill in the art would consider as a reasonable amount of deviation to the recited numeric values (i.e., having the equivalent function or result). For example, this term “about” can be construed as including a deviation of ±10 percent of the given numeric value provided such a deviation does not alter the end function or result of the value. Therefore, a value of about 1% can be construed to be a range from 0.9% to 1.1%.

[0083] The term “and / or” can mean “and” or “or”.

[0084] Unless stated otherwise herein, the articles “a” and “the”, when used to identify an element, are not intended to constitute a limitation of just one and will, instead, be understood to mean “at least one” or “one or more”.

[0085] Aspects of the system and method disclosed herein provide an advanced Al- powered platform (termed ElephasCare™) that autonomously and continuously monitors the activity of one or more subjects (e.g., residents) without the use of cameras or wearable devices. In particular, an aspect of the system and method proposed herein comprises a monitoring solution that incorporates one or more radar sensors and uses electromagnetic (EM) waves to monitor residents contact-free. As will be understood, the system comprises at least one processor connected to one or more radar devices, wherein the at least one processor comprises executable code for controlling signals between the transmitters and receivers of the radar devices, and for processing the received data. In one aspect, theprocessor is configured to execute with any number of algorithms that are designed for the various data processing steps discussed herein. In some cases, the algorithms may comprise data analysis algorithms, including but not limited to deep learning (DL), machine learning (ML), Generative Al, or artificial intelligence (Al) algorithms that are designed to interpret data received from one or more radar devices or other sensors associated with the system. It will be understood that the system further comprises at least one memory component for storing received data and / or for storing one or more libraries comprising historical and / or training data that can be accessed by one or more of the algorithms. The system is not limited to any specific processor(s) or memory component(s).

[0086] In an aspect, the system is configured for communication with a server. The server may be located in the same location as the system or at a remote location. The server may also be cloud-based. The server may, in one aspect, comprise the at least one processor and the at least one memory component. Thus, in one aspect, the radar sensors of the system may be remotely located from the rest of the system components.

[0087] In an aspect, the system is configured for communication with one or more user devices, such as laptops, tablets, handheld devices, cell phones, etc., whereby signals and / or messages are transmitted by the system to such one or more user devices. As discussed further herein, such messages may comprise alerts or the like wherein information concerning one or more subjects is transmitted to a care provider or other such person. In one aspect, communication with users is accomplished using an app loaded on one or more of the user devices. As described further herein, one aspect of the system and method comprises one or more graphical user interfaces (GUIs) that are displayed on one or more user devices, which allows users to visually receive relevant data regarding one or more subjects. In one aspect, the GUI allows users to transmit information to the server and / or the system, such as an acknowledgement that the subject’s information was received or that a response to an alert is underway. Examples of such GUIs are provided herein for illustrative purposes only. It will be understood that that the system and method can be designed to incorporate any form of user interaction methodologies, including voice, text messaging, etc.

[0088] In one aspect, the method and system disclosed herein use wireless transceivers, for example millimeter (mm)-wave radar transmitters; generally operating at frequencies between 30 GHz to 300 GHz). In addition to providing higher accuracy, mm- waves are non-ionizing, unlike x-rays which do sort ionize body molecules. Mm-waves are also absorbed by water and do not penetrate a subject’s body. Electromagnetic absorptionat these frequencies is more than 10 times less than those used for cellular bands and WiFi transmissions. Furthermore, for other applications requiring longer range or the ability to sense through obstacles such as walls — enabling coverage of multiple rooms with a single sensor — lower frequencies in the range of 1-30 GHz can also be employed. These lower frequencies offer extended coverage and enhanced penetration capabilities, making them suitable for scenarios where broader area monitoring and through-wall sensing are necessary, albeit with potential trade-offs in resolution and accuracy compared to higher mm-wave frequencies.

[0089] The transceiver will have one or more transmitters sending a sequence of EM signals. The signals will be reflected and scattered by an object (for example, an object in the environment, such as a human, as shown in Fig. 1). The transceiver will also have one or more receivers that receive signals reflected and scattered from the object. The system includes one or more processors that comprise one or more executable signal processing algorithms that are applied to the received signals in order to identify the object and / or differentiate between various objects and / or determine characteristics of the object, such as life signs (e.g., an individual’s heartbeat, breathing, or the like) or position / movement of a subject. It will be understood that depending on the radar bandwidth and machine learning / signal processing algorithms involved, objects may be in a range of distances from the system. In some cases, the objects may be between a few millimeters to several meters from the system / transceiver. In some cases, there may be more than one object that is analyzed / sensed / monitored.

[0090] Radar, short for “radio detection and ranging”, has traditionally been used to detect the location and movement of objects. This is achieved through the emission of electromagnetic radiation from a transmitter and the detection and measurement of the waves that are reflected back to a receiver. Doppler radar can be used to detect the velocity of an object that is being measured, whereas the range can be calculated based on how long it takes for the waves to be received by the sensor after they are transmitted. Depending on the angle of reflection and the degree of refraction that has occurred, it is possible to determine properties regarding the material composition of the object. Depending on the number of transmitters and receivers, angle information also could be obtained.

[0091] In one aspect, the system and method described herein provide a means of fall detection and fall prevention that goes beyond passive collection of data relating to a given subject. This data may comprise one or more of vital signs, gait analysis, sleep quality,activity record, and washroom use patterns, etc. By gathering this comprehensive information, the described system and method can provide actionable insights that aid in identifying subjects who require proactive care. In another aspect, and as described further herein, the collected data also allows for the detection of early warning signs related to various health conditions such as urinary tract infections (UTIs), pneumonia, stroke, heart failure, and more. By analyzing the data patterns and identifying deviations from normal parameters, the present system and method provides alerts and notifications to caregivers, enabling timely intervention and appropriate care for the subject. Through this proactive approach to health monitoring and analysis, the system and method described herein aims to improve overall resident well-being and to enhance the quality of care provided. By leveraging the power of data and actionable insights, caregivers can effectively prioritize their efforts and deliver targeted care to those residents who need it the most.

[0092] To effectively monitor residents within a facility, comprehensive electromagnetic wave coverage of the entire environment is needed, as illustrated in Fig. 2. A detailed investigation into determining electromagnetic coverage within a room is depicted in Figs. 3(a) and 3(b). This determination could also be conducted through simulations employing electromagnetic software tools such as HFSS (High-Frequency Structure Simulator), CST (Computer Simulation Technology) or other available software.

[0093] Depending on the size of the environment, the number of rooms to be covered, the type of radar sensors, the number of antennas, and the desired features, multiple radar sensors may be employed. The positioning of each sensor may be determined in advance, prior to installation. For example, each radar system comes with its own set of limitations, such as maximum detection range, output power, gain etc. Additionally, the regulating bodies may impose restrictions on the output power for indoor radars, thereby placing a limit on the maximum coverage range. Consequently, in larger environments or rooms, multiple radar transmitters may be necessary to ensure comprehensive coverage. For specific functionalities, such as monitoring of a subject’s breathing or heart rate, particularly when in a specific location, such as in bed, a dedicated radar for bed monitoring and another for activity recognition might be required in addition to radars for monitoring other activities within the room. Furthermore, to cover specific areas like the living room, washroom, and bedroom, due to the constrained coverage range and lower penetration rate in the mm-wave band, it may be necessary to install multiple sensors in each environment.

[0094] The placement of radar sensors within an environment may be adjusted to optimize their functionality for specific applications. Sensors can be installed in various locations, such as on walls, beneath beds, embedded in furniture or behind chairs, and mounted on ceilings, to suit the unique requirements of each monitoring task. For instance, detecting vital signs — a task that demands high precision — may render wall-mounted sensors less effective due to their distance from the subject. In such cases, placing radar sensors under the bed or behind chairs, closer to the subject, enhances detection accuracy. Additionally, for applications like bed exit attempt detection or distinguishing between routine movements and those intended to prevent pressure ulcers, a more complex setup may be necessary. Here, integrating a radar sensor under the bed to work in tandem with a wall- mounted sensor allows for a more nuanced analysis. By correlating data from both positions, the system can accurately identify the nature of the movement, differentiating between a simple repositioning and a potential bed exit attempt. This multi-location approach not only increases the system's sensitivity to various activities but also significantly improves its ability to deliver precise and contextually relevant data, crucial for applications like patient monitoring in healthcare settings. To achieve superior coverage and heightened sensing precision, deploying multiple radar sensors in varied configurations — specifically bistatic and multistatic modes — could be imperative. In bistatic mode, the radar system employs separate locations for the transmitter and receiver, leveraging the spatial diversity to detect movements or vital signs with increased accuracy, especially in obstructed environments. This configuration can significantly enhance the detection capabilities by providing different perspectives of the monitored area, thus reducing dead zones and improving the reliability of the data collected. Conversely, in multistatic mode, where multiple transmitters and receivers are utilized, the system gains the advantage of comprehensive coverage from multiple angles. This not only aids in creating a more detailed and accurate representation of the environment but also ensures the robust detection of subtle changes or activities, critical for applications requiring fine-grained analysis such as sophisticated motion detection or detailed health monitoring. The strategic deployment of sensors in these modes allows for a scalable and adaptable monitoring solution, capable of accommodating the diverse and dynamic nature of real-world environments.

[0095] As illustrated in Fig. 32, the customization of radar sensors with supplementary components — such as rod antennas or dielectric lenses — serves as a pivotal enhancement for targeted applications, optimizing sensor functionality. Rod antennas can be integrated to extend the sensor's range and improve its directional sensitivity, making it particularly usefulfor monitoring large areas or detecting movements from a distance. On the other hand, dielectric lenses focus the radar's signal, enhancing its resolution and enabling the precise detection of minute movements, crucial for applications like vital sign monitoring or the intricate assessment of sleep patterns. These add-ons not only bolster the sensor's inherent capabilities but also tailor its performance to meet the specific demands of various monitoring tasks. By leveraging these enhancements, the radar system can achieve a higher level of accuracy and sensitivity, ensuring that even the most subtle phenomena are detected and accurately interpreted, thereby significantly extending the utility and applicability of radar technology in diverse fields such as healthcare, security, and ambient assisted living.

[0096] Beamforming techniques, alongside the use of phased array antennas, stand as pivotal methods for enhancing the radar system's directional capabilities, enabling precise beam steering towards specific targets or areas within an environment. This targeted approach facilitates the focusing of the radar's energy on particular individuals, thereby optimizing the system for tasks such as vital signs monitoring, counting occupants, recognizing activities, or identifying individuals. Through the dynamic adjustment of the antenna elements' phase and amplitude, beamforming allows for the radar beam to be directed and reshaped in real-time, catering to the continuously changing conditions within the monitored space. This adaptability could be crucial for accurately tracking multiple people simultaneously, distinguishing between different types of movements, and ensuring the reliable collection of data across various scenarios. By harnessing these advanced techniques, the radar system significantly enhances its utility across a broad spectrum of applications, providing a versatile and effective solution for both security and healthcare monitoring purposes.

[0097] In a study aimed at achieving maximum coverage, two radar sensors were installed on two walls, each wall being positioned at a 90-degree angle relative to the other, as illustrated in Fig. 4.

[0098] In a separate study (see above), a single sensor was strategically placed in the main room to oversee the living area and monitor its occupants (Fig. 5 (a)). Additionally, another sensor was installed in a bedroom to specifically monitor residents within that space (Fig. 5 (b)), and a third sensor was placed in a washroom to cover its designated area (Fig. 5 (c)).

[0099] In larger facilities, such as long-term care homes or retirement homes, all sensors can be connected to a local server for on-site data processing. Alternatively, the data may be transmitted to the cloud for remote processing. In one aspect, as depicted in Fig. 6, a discrete, private, physical network is employed to stream the data, ensuring immunity to bandwidth contention and enhancing overall security.

[0100] The presently described system may utilize any radar system although, as described herein, some limitations may be imposed on the components of the system, such as external limitations from regulatory bodies that may stipulate the power levels that may be used in interior spaces. The following table summarizes examples of known radar systems (described further below) that may be incorporated into the presently described system.

[0101] It will be understood that, depending on the type of radar used, and the number of transmitters and receivers incorporated into the system, comprehensive information about objects, including their elevation and other positional details, can be provided. In this regard, and as would be known to persons skilled in the art, each radar type necessitates specific signal processing techniques. For example, Frequency Modulated Continuous Wave (FMCW) Radar delivers accurate range and velocity measurements with high resolution. Ultrawideband (UWB) Radar provides precision in range and imaging, suitable for applications like ground-penetrating radar and medical imaging. Continuous Wave (CW) Radar is utilized for ongoing target velocity monitoring, relying on the Doppler effect. Additionally, the number of transmitters and receivers influences information provision.

[0102] For example, an FMCW radar system with a single transmitter and receiver lacks the capability to provide angle information. However, an FMCW system would be preferable for the purpose of monitoring vital signs of a subject owing to its greater precision.

[0103] The choice of a transmitter may be limited by regulating authorities, such as the FCC, which may prescribe the allowable bandwidth for the presently described applications, thus also influencing the selection process for the components of the system. For example, aside from potential hardware limitations that could restrict support for extremely widebandwidths, the FCC imposes a restriction on indoor applications, limiting them to frequencies of 500 MHz

[0104] System Configuration

[0105] As noted above, radar-based systems for monitoring patients and the like are known. However, such known systems rely on expensive radar sensors (i.e., transmitter and receiver devices) that are often designed for locally processing signals (e.g. EDGE devices) and / or for transmitting processed data over wireless networks. However, such known devices and methods are costly and are limited with respect to traffic over wireless networks. Thus, implementing such a system throughout a large facility, such as a long-term care facility with many rooms would be unfeasible. The present system, on the other hand, provides a unique solution to such problem by utilizing a simplified radar sensor that is connected directly to server over a LAN or other such wired connection. It will be appreciated that although the system is uniquely designed for implementation over a wired network, such configuration is designed to overcome deficiencies in currently known wireless networks. Accordingly, it will be understood that the present system may be implemented over any data network.

[0106] To address the above-noted issues, the present inventors developed a unique radar sensor system, as illustrated in Fig. 6. In one example, a number of model BGT60TR13C 60 GHz radar sensors (Infineon Technologies AG) were used. The core functionality of BGT60TR13C is to transmit frequency modulated continuous wave (FMCW) signal via one of the transmitter channels (TX) and receive the echo signals from the target object on the three receiving channels (RX). Each receiver path includes a baseband filtering, a VGA, as well as an ADC. The digitized output is stored in a FIFO. The data are transferred to an external host, microcontroller unit (MCU) or application processor (AP), to run radar signal processing. It will be understood that the present description is not limited to any particular radar sensor. However, in view of the design of the present system, the radar sensors do not need to be sophisticated, thereby allowing simple known devices to be used.

[0107] The sensors are connected directly to a server over a wired (LAN) connection in a “chip to chip” manner. In this way, the raw data from the radar sensors are transmitted directly to the server where, as discussed further below, the data is then processed through one or more algorithms programmed in the server to derive the desired patient / resident information. As also illustrated in Fig. 6, data processed by the facility server may, in oneaspect, be transmitted to an onsite or offsite server of a service provider, where such data may be further processed and / or analyzed for storage, report generation, etc. The third party service provider may also forward such information as may be required to devices of authorized personnel. In one example, such information may comprise alerts or the like in reference to one or more patients or residents under care.

[0108] Currently, no known radar-based system comprises such direct sensor to server connection for transmitting raw data. As would be appreciated, with the presently described system, scale up to include additional sensors if facilitated.

[0109] Furthermore, by using the sensors to simply provide raw, unprocessed data, it will be understood that the server of the system may be programmed to automatically execute one or more algorithms that are specifically designed to process the raw data. As discussed herein, such algorithms may process the raw data to determine the presence or absence of a person in a room, the activity of the person, to determine falls, etc. Such flexibility is realized as a result of raw data being collected at the server.

[0110] In one implementation of the present system, the radar sensors, comprising printed circuit boards (PCBs), were mounted on walls at a height of about 230 cm, or about 7.5 feet, from the floor. Moreover, the sensors were mounted so that PCBs were oriented at an angle of 30 degrees from vertical for beam delivery, which was determined to offer maximum coverage in an area. For this purpose, a unique housing was designed for accommodating and mounting the radar sensors in the desired orientation. An example of such housing is illustrated in Figs. 34 and 35. As shown, the housing 3410 comprises a rear wall plate 3412 that is adapted for mounting to a wall. For this purpose, the wall plate 3412 may be provided with apertures to accommodate fasteners or other such devices to affix the wall plate to a wall. It will be understood that that the wall plate 3412 may be attached to a wall in any manner. The housing 3410 further comprises a cover 3414 that is adapted to attached to the wall plate 3412, such attachment being of any known manner. The cover 3414 is adapted to contain the aforementioned radar sensor PCB and is sized accordingly. Further the housing 3410 and / or the housing cover 3414 is adapted to secure the radar sensor PCB at a desired angle with respect to the wall. In one aspect, as mentioned, above, it was determined that an angle, 9 (as illustrated in Fig. 34), of 30 degrees from vertical (i.e., angled downward from the wall) is ideal for maximizing coverage. The housing cover 3414 is preferably made of a plastic material, or other material that would not interfere with the radar signal transmission or reception. In one aspect, the cover may be made of a plastic havingwalls of 3mm thickness. However, the front face 3416 of the cover 3414 is provided with a thickness of 1 mm, which was chosen to minimize interference of the cover with the radiation pattern of the radar sensor.

[0111] As will be appreciated, the present system employs radar transmitters and receivers to provide a cost-effective and contact-less monitoring of subjects. Further, by omitting the need for cameras and the like, the description preserves the privacy of the subject being monitored.

[0112] As a single-input-multiple-output (SIMO) FMCW radar, and as described above, the presently described sensors provide range, azimuth, and elevation information from the radar signals. A flow chart of the algorithm is shown in Fig. 7.

[0113] Fig. 7 illustrates a data processing algorithm used by the present system and method. Using an FMCW radar, each Correct raw data acquisition as the first step is ensured by setting the start time and sampling frequency of the Analog to Digital Converter (ADC) so that the required Nyquist samples are equally distributed within ramp start and end. For each frame, a three-dimensional data cube <t> e CNSXNCXNRX containing the complex-valued baseband signals is obtained. The first dimension, CNS, contains all samples per chirp (fast-time) for range estimation, the second dimension, Nc, belongs to the different chirps per frame (slow-time) for velocity estimation, and the third dimension, NRX, corresponds to the number of receiving antennas, where NRX= 2 RX antennas for angle of arrival (AoA) estimation.

[0114] In the initial step, a real-time process is employed where the average of the time domain signal is subtracted from each range bin for every receiver. This operation effectively eliminates any DC bias present in the signal. Subsequently, radar cubes are created, containing fast time and slow time data specific to each channel. Each radar cube is represented as an NXKXL matrix, where N denotes the number of chirps in a frame, K represents the number of chirp samples, and L indicates the number of channels. The value of L is determined by the multiplication of NTX(number of transmitters) and NRX(number of receivers).

[0115] In the second step, a Fast-Fourier Transform (FFT) is applied to the fast-time dimension with a zero padding of factor two. The resulting FFT is called range FFT. Subsequently, coherent pulse integration along the slow-time dimension aims to improve thesignal-to-noise ratio and moving target indicator (MTI) filtering is used for clutter suppression.

[0116] To extract the range-profile of one or more living subjects, the first Fast Fourier Transform (FFT), referred to as the Range-FFT, is applied to the received chirp samples. This process enables the analysis of the signal characteristics in the range domain. Figures 15a-15h illustrate this step, highlighting that the received signals consist of reflections from both stationary targets (clutter) and living bodies (residents).

[0117] The targets’ velocities are estimated by the so-called Doppler FFT along the slow-time dimension for the corresponding range bins. Zero padding with a factor of two is also applied to this FFT resulting in doubling the size of the Doppler FFT. Computing the Doppler FFT solely for the target indices reduces the computational complexity with regard to the implementation. This range-Doppler processing is done for both antennas.

[0118] TX-to-RX leakage is a common challenge in radar systems and due to the bandpass filtering in the baseband, this leakage leads to a low-frequency component, which results in high amplitudes in the first range FFT bins. TX-to-RX leakage is also encountered with the BGT24MTR12 radar chip of the Infineon radar sensors that were used. Next to extensive hardware-related cancelation methods, there are also software-based approaches to remove this leakage. One possible software solution is to calibrate the system before usage measuring the system itself by placing an absorber in front of the antennas to generate a reflection-free environment.

[0119] At runtime these calibration values are subtracted from the measurement data. On the one hand, this method simultaneously extinguishes other impairments of the radar hardware, on the other hand, the calibration values are highly temperature dependent. Another software solution is MTI filtering, which is a type of in-situ calibration and therefore temperature independent.

[0120] MTI filters in principle are low-order, simple finite impulse response (FIR) designs. At each time stamp the absolute maximum value over slow time of each range bin is denoted by rijmax. The MTI filter value ti is the weighted average of this maximum value and the previous MTI filter value t,-i , with a weight of a:

[0121] In the first time stamp to is initialized with Zero. For each range bin, the previous MTI filter value is subtracted from rijmax to obtain the filtered range FFT value n.fiit:>7. tilt — lbs( / 7, max ’ — 1 ) •

[0122] This filtered value is then utilized for the subsequent target detection. MTI processing performs linear filtering which leads to a diminished signal strength for static targets while maintaining the signal strength of the moving targets. Therefore, MTI processing helps to remove completely static targets, while non-static targets like humans are retained.

[0123] After computing the range FFT for each chirp and applying the MTI filter it is required to combine the data over all chirps of a frame for subsequent target detection. One possible data combination strategy is coherent integration, which combines the phase and magnitude of the range FFT data coherently over all chirps of a frame. Coherent integration is based on the mean Rmean of the range FFT data n over NC chirps:^mc

[0124] Here, amplitude and phase of the range FFT value for chirp i are denoted by ai and <pi , respectively. This method enhances the signal-to-noise ratio (SNR), but is has to be considered that all phase values of all chirps have to be aligned, since phase misalignment leads to signal distortion. Another approach is to find and use the chirp i with the maximum absolute range FFT value Rmax overall NC chirps:

[0125] This approach assures that the FFT data with highest amplitudes is chosen for target detection and prevents phase misalignments.

[0126] The basic approach to determine the AoA is to evaluate the phase difference between both RX antenna beams, also referred to as phase-comparison monopoles. The AoA 9 is calculated by geometrical considerations based on an incoming plane wavefront. Various algorithms were developed to calculate azimuth and elevation information of the environment. We used a Capon beamformer and refer to the document called “Radar Background”.

[0127] For some types of radar, for example, in the case of FMCW radar, as mentioned above, range-FFT (fast Fourier transform) and Doppler-FFT may be employed to derive range and velocity information. The capon beamformer can be utilized for angle information, and integration over channels or frames / time can enhance the signal-to-noise ratio. As noted above, techniques such as MTI (Moving Target Indicator) or clutter removal algorithms may be applied to eliminate static objects.

[0128] In one aspect, the innovative approach proposed herein capitalizes on the prowess of machine learning / deep learning, Generative Al and other techniques to process radar data directly, eliminating the need for traditional preprocessing and radar signal processing. By leveraging raw radar data as input, this method facilitates the training of sophisticated machine learning models, including deep learning, Generative Al, and other advanced algorithms. The training process is enriched by a diverse and comprehensive dataset comprised of numerous samples and data captured over time, alongside invaluable feedback from caregivers. This extensive dataset ensures that the models learn to accurately interpret the nuanced patterns and characteristics of the radar signals, reflecting real-world scenarios. As the models are trained to directly handle raw data, the conventional steps involved in radar signal processing become obsolete, streamlining the data analysis pipeline. The deployment of these Al models on GPUs further enhances the system's efficiency, offering rapid processing capabilities. This approach not only simplifies the implementation but also significantly improves the speed and accuracy of data analysis, paving the way for a more effective and responsive monitoring system.

[0129] Presence-Absence Detection (PAD)

[0130] In an aspect of the present description, a Presence-Absence Detection (PAD) algorithm was developed for analyzing the raw data supplied by the radar sensors for identifying the presence of individuals in a space, such as a bedroom, washroom, or otherliving quarters. It will be understood that the algorithm discussed herein are no limited to any particular living space per se and may be used also for hallways or the like.

[0131] The main objective of this algorithm is to determine which rooms in the residence are occupied. The algorithm thereby gathers information about a subject's location, such as, but not limited to, whether they are in their living area, in the washroom, the duration of washroom usage, the frequency of washroom visits, and whether they are out of their room.

[0132] To achieve this, the PAD algorithm is programmed into the server and is applied to the raw data obtained from the radar sensors. Fig. 8 illustrates an overview of the signal processing procedure involved in the PAD algorithm. This algorithm effectively differentiates between an occupied environment and a vacant one, allowing for accurately identifying the presence or absence of individuals in the space being monitored.

[0133] In an initial step, a real-time process is employed where the average of the time domain signal is subtracted from each range bin for every receiver. This operation effectively eliminates any DC bias present in the signal. Subsequently, radar cubes are created, containing fast time and slow time data specific to each channel. Each radar cube is represented as an NxKx|_ matrix, where N denotes the number of chirps in a frame, K represents the number of chirp samples, and L indicates the number of channels. The value of L is determined by the multiplication of Ntx(number of transmitters) and Nn< (number of receivers).

[0134] As illustrated in Fig. 8, once the Range-Profile of subjects for each channel is created without clutter, the next step is to detect the presence of individuals. To achieve this, the present description leverages one of the unique characteristics of human subjects, which is their continuous breathing motion. Human breathing manifests itself in the micro-Doppler signature as consistent chest motion over time.

[0135] It will be understood that the data from the sensors is not used to extract an exact breathing waveform or breathing rate, as that would be a complex task requiring knowledge of the appropriate range of chest motion specific to each individual. Additionally, accounting for the entire body's motion and the signals generated at different ranges further complicates the process. Further, as would be understood, extracting the precise breathing waveform would introduce more complexity to the present signal processing chain, whereas the objective in the present description is to develop a simple, accurate, and efficient algorithm.

[0136] In line with these objectives, the PAD algorithm utilizes the following observations regarding the effects of breathing on the received signals over time:

[0137] (1) The chest movement associated with breathing remains consistent over time.This consistency is reflected in the micro-Doppler pattern of the received signals from humans, which exhibit a high degree of correlation over time. This correlation serves as a valuable characteristic for detecting the presence of individuals in the residence.

[0138] (2) Even if the resident moves within the monitored area, the micro-Doppler signals originating from the chest movement still exhibit correlation over time. This observation allows us to overcome the challenge of subject movement while maintaining the effectiveness of our algorithm in detecting the presence of individuals.

[0139] By leveraging these two observations, the algorithm takes advantage of the consistent micro-Doppler patterns associated with human breathing to identify the presence of residents accurately and efficiently in the monitored area.

[0140] To demonstrate the impact of chest movements on the received signals of each channel, we apply the second Fast Fourier T ransform (FFT) in the slow time domain to the previously obtained range profile. This process allows us to analyze the effects of micro- Doppler patterns. It is important to note that the precise position of the subject is not crucial for detection purposes.

[0141] To enhance the signal intensity and improve detection performance, we perform coherent accumulation first on the range bins (fast time) and then on the channels (across receivers). The PAD algorithm proceeds by determining the presence of a resident based on the correlation coefficient between two consecutive observation signals (OSs), which represents the integration of the Doppler-FFT information over range, receivers, and NF frames, denoted as correlation (OSj, OSj+i). If the correlation coefficient exceeds a predefined threshold, the presence of a residence is identified.

[0142] The outcome of this step is the integration of the Doppler information over both range bins and channels, resulting in a parameter referred to as lnt_Dopp. This integration significantly enhances the signal intensity and Signal-to-Noise Ratio (SNR). However, since our residents typically exhibit minimal motion and sometimes have a low breathing rate, performing clutter removal might inadvertently remove the subject from the radar signal when they are not actively breathing.

[0143] To address this issue and further improve the detection performance, we integrate the lnt_Dopp from NF (a specified number of) frames with each other. This integration process generates an observation signal (OS) which represents the integration of the Doppler-FFT information over range, receivers, and NF frames. This integration allows us to capture and preserve the subject's presence even during periods when they are not actively breathing. lnt_Dopp i = TF(Tj, f),Tj = i x Ts, i: (NOsi— 1) x NO + 1 to NOsix NO.

[0144] Finally, OSj is calculated as:

[0145] The PAD algorithm proceeds by determining the presence of a resident based on the correlation coefficient between two consecutive observation signals (OSs), denoted as corr(OSj, OSj+i). If the correlation coefficient exceeds a predefined threshold, denoted as thr, the presence of a residence is identified.

[0146] To store the results of the PAD algorithm, we maintain a storage capacity of NO times. This means that if the PAD algorithm identifies the presence of a residence more than N_det times, the presence status will be recorded in our database. Therefore, to accurately determine the resident status, we require a total of NO multiplied by the number of frames ( / VF).

[0147] In an aspect of the present study, the application of machine learning and deep learning models has been shown in distinguishing between vacant and occupied rooms, as well as confirming the presence or absence of individuals. This advancement significantly departs from conventional methods by negating the necessity for intricate signal processing techniques that traditionally required the definition of specific thresholds to interpret radar data. Instead, these intelligent models learn to discern patterns directly from the data, effectively identifying occupancy status without the need for pre-defined criteria. This shiftnot only streamlines the detection process but also enhances accuracy and reliability, as it leverages the comprehensive learning capabilities of ML and DL algorithms. By training these models on diverse datasets that capture a wide range of scenarios, they adapt to recognize subtle nuances in the radar signals, thus eliminating the limitations associated with manual threshold settings and complex signal processing. This approach not only simplifies the operational framework but also opens up new possibilities for more nuanced and responsive environmental monitoring.

[0148] Resident Localization

[0149] With the azimuth information of the target obtained from the capon beamformer and the range calculated from the range-FFT, a range-azimuth heatmap of the subject is obtained. The range-azimuth heat map represents the density of reflected signals in the environment. Due to the low angular and range resolution of the radar sensor, obtaining the accurate relative position of the resident is challenging. To obtain the accurate position, a novel algorithm, referred to herein as “get_position” was developed. The details of the algorithm are provided in Fig. 9.

[0150] As illustrated in Fig. 9, to derive the range, azimuth, and elevation information, a coherent integration process is employed on the Range-Doppler Map (RDM) over Doppler, Range-Azimuth Map (RAM) over range, and Range-Elevation Map (REM) over range. This integration allows for the generation of a range (R) vector, azimuth (Az) vector, and elevation (El) vector.

[0151] First, coherent integration is performed on the RDM over the Doppler dimension, which combines the information from multiple Doppler profiles. This integration enhances the detection and estimation of the target’s range.

[0152] Next, coherent integration is applied to the RAM over the range dimension. This integration process combines the azimuth information from multiple range bins, resulting in an azimuth vector that provides the angular position of the target in the azimuth plane. Similarly, coherent integration is performed on the REM over the range dimension to combine elevation information from multiple range bins. This integration yields an elevation vector, which indicates the vertical angle of the target. By performing coherent integration on the RDM, RAM, and REM, we obtain the range, azimuth, and elevation vectors, respectively. These vectors provide comprehensive information about the target’s spatial position, allowing for accurate localization and tracking in three-dimensional space.

[0153] After obtaining the range, azimuth, and elevation vectors through coherent integration, a 1-D CFAR (Constant False Alarm Rate) detection process is applied to each vector. This CFAR algorithm is employed to identify and detect points of interest in terms of range, azimuth, and elevation. Once the points of interest are detected, a 1-D clustering technique is used to cluster these points based on their proximity in range, azimuth, and elevation. This clustering process groups together the detected points that likely correspond to the same subject or target. To obtain an accurate position of the subject, the average of each cluster is considered as the position of the subject. By calculating the average of the clustered points in range, azimuth, and elevation, the system determines the precise position of the subject in three-dimensional space.

[0154] The combination of the 1-D CFAR algorithm and 1-D clustering approach allows for the accurate identification and localization of the subject by refining the detected points and estimating their positions. This methodology enhances the precision of the subject’s position estimation based on the detected points in range, azimuth, and elevation.

[0155] Building upon the foundation laid by conventional techniques for mapping and locating subjects, the integration of Generative Al may be used as a transformative approach to accurately determining the position of the subject within an environment. Leveraging the capabilities of Generative Al, this method transcends traditional limitations by utilizing advanced algorithms that can interpret and enhance the range-azimuth heat maps generated from radar data. This approach facilitates a deeper understanding and refinement of the subject's position, overcoming the challenges posed by the radar sensor's inherent low angular and range resolution. By employing Generative Al models, which are trained on extensive datasets encompassing a wide array of environmental conditions and subject positions, this innovative method achieves unprecedented accuracy in real-time location tracking. The Generative Al models dynamically generate and refine predictive models of the environment, enabling them to pinpoint the subject's position with a level of precision previously unattainable through conventional means alone.

[0156] In this regard, despite the low range and angular resolution of the sensors used in our study, we obtained highly accurate results relating to the position of a subject after processing the raw radar data with our novel algorithm, described in Fig. 9, which is based on unsupervised machine learning (clustering). For example, in one test, the range-doppler maps of two consecutive frames, and range-azimuth and range-elevation maps of a frame, based on low range resolution (30 cm) of the radar data, provided a range of a sitting subjectfrom 2 m to 4 m, and angle information that was also extended in range and angles. However, when the raw data was processed with the clustering algorithm described herein, the precise location of the subject was identified. Thus, we have demonstrated that our unsupervised based positioning algorithm overcomes the low range and low angular resolution of the radar sensors we used and accurately locates the position of a subject. Similarly, the get_position function mentioned above can also serve to provide in-bed status information (as discussed further below).

[0157] Fall Detection

[0158] In one aspect, to enable prompt detection of fall incidents, the system described herein may comprise one or more algorithms, or functions, that serve to improve the accuracy of identifying the type and potential severity of falls while minimizing false positives. These algorithms are generally illustrated in Fig. 10 and are identified as: (i) Suspected Fall Detection (SFD) Algorithm; (ii) Activity Recognition (Al / Deep Learning Analysis); (iii) Position Monitoring (“get_position” function); and (iv) In-Bed Status Check. These aspects are described further below.

[0159] (i) Suspected Fall Detection (SFD) Algorithm.

[0160] As illustrated in Fig. 11a, this algorithm analyzes the data from the radar sensors for detecting sharp or abrupt changes in speed, which can be indicative of a fall. However, since other activities may also produce similar speed changes (e.g., fast movements, jumping into bed, rapid transitions between sitting and lying down), additional functions are activated to validate or dismiss the occurrence of a suspected fall.

[0161] As shown in Fig. 11a, to obtain the Doppler information, a coherent accumulation is conducted over RDM in the range dimension. This accumulation process enhances the Doppler signal and results in a Doppler vector containing velocity information. Since falls typically involve higher speeds, a low-pass filter is applied to the Doppler vector to isolate the velocity range associated with falls. This filtered output is referred to as F_Dopp, which specifically captures velocities relevant to fall incidents.

[0162] To further validate the occurrence of a fall, the correlation between two consecutive F_Dopp vectors is calculated. If the correlation exceeds a predefined threshold, it indicates a consistent pattern of Doppler changes over time, which is characteristic of a fall event. Upon detecting a significant correlation, a CFAR is employed to identify the detectedspeeds associated with the fall. CFAR examines the F_Dopp vector and identifies points of interest corresponding to potential fall velocities. If CFAR detects a sufficient number of points that satisfy the detection threshold, a suspected fall alert is triggered. This alert prompts the activation of other aforementioned functions to confirm or dismiss the fall incident. These functions include deep learning analysis, position monitoring, and in-bed status checks.

[0163] By incorporating coherent accumulation, low-pass filtering, correlation analysis, and CFAR, the present system accurately identifies potential fall incidents based on the Doppler velocity information. This approach reduces false positives and ensures that a comprehensive evaluation is performed before confirming or dismissing a fall alert.

[0164] (ii) Activity Recognition - Al / Deep Learning Analysis.

[0165] When the SFD algorithm identifies a potential fall, the system considers the results from a deep learning model to determine if the subject is actually walking or engaged in a fall-like behavior. This algorithm is illustrated in Fig. 12. By leveraging deep learning capabilities, the system enhances the accuracy of fall detection by considering the subject’s overall movement patterns.

[0166] As shown in Fig. 12, the proposed algorithm consists of two processes: (1) walking period identification and activity recognition and (2) gait parameter extraction. In our proposed system, raw data from the radar is collected from a radar. There are two types of clutter effects in received signals: (1) stationary clutter and (2) time-varying clutter (ghosts). The direct reflection from stationery or unanimated objects is called stationary clutter. The stationary clutter removal algorithm is applied to the range-doppler profile to remove signals reflected from stationary clutter. However, the interaction between a subject and a stationary object creates multipath or ghosting effects. After performing the stationary clutter removal algorithm, the remaining signals in the range profile are direct signals from the subject, caused by chest motions (breathing) and other motions created by performing in-home activities in addition to multipath effects. It is demonstrated that deep learning can classify in- home activities despite the existence of multipath effects or ghosts. Since the human body is non-rigid, reflections from a human body occupy multiple cells of range bins in the range profile. Human locomotion, including walking, is a complex motion, and the velocity of each segment of the human body performing different tasks varies over time, producing various micro-Doppler shifts in scattered signals. Applying the second FFT on a series of radarchirps (i.e., frame), a range-Doppler map (RDM) is obtained. Therefore, using an FMCW radar, we simultaneously provide a range-Doppler map at each frame containing range and micro-Doppler signatures of a subject’s in-home activities.

[0167] Given that our target is a single subject, we use the entire RDM to train the model. This simplicity helps us avoid other signal processing such as detection (to capture occupied bins), clustering methods to cluster the detected bins, and then association algorithms to associate new bins to the previously occupied bins. Any in-home gait extraction method is prone to failure if the system is not intelligent enough to identify a human’s in-home activities and differentiate between them. The system provides precise and accurate gait data and be able to track a subject’s in-home activities over long periods. The system is able to identify the type of in-home activities a subject performs. Five classes are defined in this study: (1) “Empty”, (2) “Sedentary- out of bed”, (3) “In-bed”, (4) “Active_ in- room”, and (5) “Walking”. These are some of the activities a subject performs during a typical day.

[0168] Owing to the complexity of human motion, complex signal processing is required to map the RDM patterns to a human’s specific activity, which is mathematically not feasible. For this reason, we have adopted machine learning as an effective tool for our system. During the development of the system, we considered several machine learning and deep learning approaches for activity recognition, such as support vector machines, random forests, and convolutional neural networks. We evaluated the performance of these approaches using various metrics such as accuracy, precision, and recall and compared them to determine the best approach for our specific application. Conventional machine learning algorithms are limited in their capacity to fully capture the rich information contained in complex data, particularly time-varying samples. Our proposed system in this study leverages deep learning approaches to use the resulting time-varying signatures of the subject being monitored. Using multiple deep layers in a single network enables the efficient extraction of a subject’s features and the building of a classification boundary.

[0169] Many deep learning models have shown exceptional promise in radar-based human activity recognition systems. The raw data is commonly converted into a 2D spectrogram using the STFT method while being treated as an optical image. The corresponding architectures, such as 2D convolutional neural networks (2D-CNNs), are used in these systems. However, since a human body motion consists of a series of associatedpostures through time, ignoring these temporal characteristics could lead to a complex network with many parameters, which may not result in accurate recognition.

[0170] Deep recurrent neural networks (DRNN) have successfully addressed classification problems that feature temporal sequences. DRNNs use a hidden node as memory, passing previous information to the next state for processing sequential inputs. Through this process, a DRNN can extract the temporal features of data. Long short-term memory (LSTM) and gated recurrent unit (GRU) are the two common models for sequential learning. Due to the complex structure of a single LSTM unit, the LSTM network contains many parameters and so requires a larger sample size. LSTM contains three gates: the forget gate, the input gate, and the output gate. On the contrary, a GRU network has a simpler structure and fewer parameters. A GRU network includes only the reset gate and the update gate. From a spatial complexity perspective, LSTM has more parameters than GRU, therefore, GRU has fewer computation costs than LSTM.

[0171] We realized that the RDM has enough features for a single subject in-home monitoring, and the Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks are promising models to be used for time-varying RDMs of human activity classification. We demonstrate that GRU achieves sufficient recognition accuracy with relatively low complexity without the need for the subject’s point cloud information. The advantages of RDM, compared to point cloud information, are that such a system can provide valuable information using only one transmitter and a receiver. Additionally, preprocessing is faster and simpler. An alternative approach using point cloud information would require an expensive high-resolution radar and complex signal processing.

[0172] (iii) Position Monitoring.

[0173] As falls often result in changes in the subject’s position, the system employs a “get_position” function (Fig. 9) to accurately track the subject’s position. By monitoring any significant changes in the subject’s position, the system and method further validate the occurrence of a fall.

[0174] (iv) In-Bed Status Check.

[0175] The deep learning model also assesses whether the subject is in bed or not. If the model does not detect the subject in bed at the time of the suspected fall, it indicates a higher likelihood of an actual fall incident. In such cases, the fall detection is confirmed.However, if the subject is recognized as being in bed, the fall incident may be dismissed as a false positive.

[0176] By integrating one or more of the aforementioned algorithms and functions, the system and method described herein achieves a robust fall detection system that combines various cues, such as changes in speed, deep learning analysis, position monitoring, and inbed status checks. This integrated approach enhances the accuracy of fall detection while minimizing false positives.

[0177] System Modifications and Implementations

[0178] As depicted in Fig. 13, and as discussed below, the present system and method allows for all signal processing algorithms to be replaced with Al models. In addition to fall detection, an Al-powered platform for resident activity tracking may be employed to proactively prevent falls and identify signs of illness. For example, the system can prevent falls either by predicting issues through data analysis or by direct detection methods. Additionally, the system analyzes gait over time, detecting any degradation beyond predefined thresholds for gait patterns, which helps predict future falls. Direct fall prevention is also achieved by detecting bed exit attempts for non-ambulatory residents with dementia. Since they may not be aware of their lost ability to walk, the system identifies attempts to exit the bed, sending alerts to caregivers for timely intervention and preventing the resident from leaving the bed.

[0179] More specifically, in developing the model illustrated in Fig. 13, the following steps were performed.

[0180] 1 . Data collection: Radar sensors were strategically placed in the environment where residents typically move, such as corridors, bedrooms, and bathrooms. These sensors continuously emitted radio waves and collected the reflections (echoes) from objects within their range. Data collected included information about the distance, velocity, and direction of movement of objects (including residents) within the monitored areas. Over time, a large dataset was accumulated, capturing various movement patterns, behaviors, and interactions within the environment.

[0181] 2. Model training: The collected radar data served as the training dataset for developing Al models. Various machine learning and deep learning techniques, such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs), were employedto process and analyze the radar data. Supervised learning methods were used to train models to recognize specific patterns associated with different activities, such as walking, sitting, standing, or attempting to exit the bed. Unsupervised learning methods have been utilized to discover hidden patterns or anomalies in the data that could indicate potential health issues or fall risks.

[0182] 3. Testing and Validation: The trained models were subjected to rigorous testing and validation to ensure their accuracy and reliability in real-world scenarios. The testing involved feeding the models with new radar data that were not used during the training phase. Metrics such as accuracy, precision, recall, and F1 score were calculated to evaluate the performance of the models in detecting falls, predicting health issues, and identifying abnormal behaviors. Cross-validation techniques were employed to assess the generalization ability of the models across different subsets of the dataset.

[0183] 4. Real-time Implementation: Once the models demonstrated satisfactory performance in testing, they were deployed for real-time implementation in a long-term care facility. The radar sensors continuously collected data from the environment, which were then processed in real-time by the trained Al models. As soon as a potential fall risk or health issue was detected, the system triggered appropriate alerts or notifications to caregivers for timely intervention. Integration with the existing monitoring infrastructure and protocols of the facility ensured seamless operation and coordination with other healthcare services and personnel.

[0184] As shown in Figs. 14a to 14c, the present inventors have demonstrated that the Short-Time Fourier Transform (STFT) of signals or spectrograms, or even radar raw data of activities can be used to train deep learning or machine learning models for the identification of specific activities performed by individuals.

[0185] As illustrated in Figs. 15a to 15h, the present inventors have demonstrated that a series of range-Doppler maps of activities may be employed to train deep learning or machine learning models for the identification of specific activities performed by individuals.

[0186] As shown in Figs. 16a to 16d, a distinct contrast is evident between an empty room and a person in bed when examining the range-Doppler maps.

[0187] As depicted in Fig. 17, certain studies employed range-azimuth, range-elevation, and range-Doppler maps separately for training models.

[0188] In a study involving seven subjects for training and testing deep learning models, both Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks were identified as more accurate and faster for in-home activity recognition. The GRU network was chosen for real-time processing due to its speed in predicting new classes, nearly matching the accuracy of the LSTM network in predicting in-home activities for new subjects. Performance graphs illustrating the accuracy and loss function of the GRU network are presented in Fig. 18.

[0189] We provide the outcome of GRU as it yields the best performance. The confusion matrix of the GRU network is provided in Fig. 19.

[0190] In another study, gait patterns of 10 subjects were analyzed, as illustrated in Fig. 20. Parameters such as the position of a walking person (Fig. 21), torso velocity (Fig. 22), number of steps and step points (Fig. 23), step time, and walking speed were obtained.

[0191] In certain aspects, the system and method disclosed herein may utilize data from multiple sources for analysis. As depicted in Fig. 24, data from radar sensors, along with information from other systems / resources, may be combined to offer additional insights through comprehensive analysis. In this regard, such other systems include various available sensors such as infrared, pressure sensors, in-bed sensors, and systems such as nurse call (or call-bell), pull cords or motion detectors, NFC access control or other security systems. Additional resources may comprise the resident’s profile, lab tests, cognitive test / scores, physical and cognitive conditions, reports from physiotherapists, geriatricians, history of falls, history of stroke, etc.

[0192] Aspects of the system and method disclosed herein leverage technologies to allow for improved resident-centred care and reduce task-intensive activities, allowing more time for direct care. As described above, aspects of the system and method disclosed herein may be used to predict and thereby prevent falls and cascading effects of undetected illnesses.

[0193] Aspects of the system and method disclosed herein track resident activity through either Al models or signal processing, which offers greater accuracy and thereby minimizes false readings.

[0194] Aspects of the system and method disclosed herein may also leverage Al models to recognize and alert for falls, recognize bed movement, to track and report on changes totypical activity. For example, in one aspect, instead of combining signal processing with machine learning and adding multiple layers for fall detection and bed movement detection, machine learning and deep learning may be trained separately for each class. For instance, various fall samples may be simulated or measured in the lab or different facilities, and similarly, for bed movement, several residents could be invited to perform bed movements. The collected data would then be fed into deep learning or machine learning models to predict future cases. This approach eliminates the entire signal processing chain mentioned above and relies solely on deep learning / machine learning.

[0195] Monitoring of Activity Variations

[0196] Aspects of the method and system disclosed herein may also use Al and Machine Learning to learn each resident’s normal activity routine and provide reporting to front line caregivers on exceptions to such individual’s normal activity. Detecting changes in activity could indicate a change in a resident’s condition. For example, in one aspect, the system establishes a unique baseline for each resident. For instance, the baseline for a resident in room 001 would typically differ from a resident in room 002. This baseline encompasses factors such as washroom utilization habits (frequency or duration), gait patterns like speed, step length, stride length, cadence, active versus sedentary hours, time spent in bed or on a chair, prolonged inactivity, and bed exit routine. Changes in these parameters may indicate underlying conditions. For example, alterations in washroom utilization could signal a urinary tract infection (UTI), while changes in gait may predict conditions like dementia, stroke, or falls. Shifts in active and sedentary hours could indicate insufficient energy, depression, or other underlying issues. The present system is capable of promptly notifying physicians and caregivers of any variations in the resident’s baseline, thereby enabling proactive intervention.

[0197] Aspects of the method and system disclosed herein could further be refined to not only monitor and report deviations from a resident's normal activity routines but also to possess predictive capabilities, potentially diagnosing specific conditions based on observed changes in the detected parameters or additional insights derived from radar data. By leveraging the power of Al and machine learning, deep learning, Generative Al and other methods, the system can analyze a wealth of data points, ranging from minute shifts in daily activities to more pronounced changes in movement patterns.

[0198] This analysis is grounded in the establishment of comprehensive baselines for each resident, incorporating a diverse array of behavioral and physiological indicators such as mobility metrics, daily routines, and interaction with the environment. In particular, the present system and method allows for the collection and analysis of a patient’s movement data over time and thereby establishes baselines for various activities. When deviations from these baselines are detected — be it in the form of altered movement patterns, changes in routine activities, or unexpected inactivity — the system utilizes sophisticated Al algorithms to correlate these changes with potential health conditions. For example, a decrease in mobility or alterations in gait might be indicative of musculoskeletal disorders or neurological conditions, whereas changes in routine bathroom visits could hint at urinary tract infections or renal health issues. Similarly, variations in sleep patterns or activity levels could suggest sleep disorders, depression, or other mental health concerns. This predictive model benefits from continuous learning, as it integrates feedback loops and constantly updates the resident's profile with new data, thereby refining its diagnostic accuracy over time. While the system does not replace medical professionals, it serves as a critical early warning tool, flagging potential health issues for further evaluation and intervention by healthcare providers. In essence, it acts as a vigilant observer, harnessing the nuanced data captured by radar sensors to offer timely insights that could lead to early diagnosis and potentially preemptive treatment of conditions, thereby enhancing the overall well-being and safety of residents.

[0199] Washroom Functions

[0200] Aspects of the method and system disclosed herein allow for the monitoring of changes in toileting routines, such as the duration or frequency of toilet visits for each resident. This data can be displayed through a graphical user interface (GUI) that is presented to a caregiver or other presonnel on a communication device, such as a mobile app or a desktop system, as shown in Fig. 25(a). In this example, the system tracks an individual’s use of the washroom where a sensor is located. The data concerning the individual’s ingress and egress from the washroom is tracked as well as the time the individual spends in the washroom. This data may then be visually represented as shown in Fig. 25(a). As noted above, the system establishes a unique baseline for each resident, whereby changes in activity with respect to such baseline can be monitored and acted upon.

[0201] The pseudo-code of an example algorithm for washroom use monitoring is provided in Algorithm 1 .Algorithm 1 : In-Home Status Recognition Algorithm Input: Radars Raw Data from each Single Board Output: Residence Status while True: chirp=capture_raw_data () room=PAD (chirp) else if room="in_washroom" save_in_washroom_date_and_time 0 else if room="out_of_home" save_out_of_home_date_time 0 else if room="Livingroom" status=check_status_of_livingroom (chirp) save status of livingroom (status, date, time)

[0202] As shown, the radar real-time raw data captured in each room is the system’s input to generate the output of the subject’s status report. If the PAD algorithm identifies the presence of the subject in the washroom (occupied washroom), the time duration of inwashroom status will be stored in the database to record the washroom usage time along with the entrance / exit time. If the PAD algorithm detects the subject in the living room, occupied living room will be stored. If the PAD algorithm identifies the absence of the subject in all areas, the status of out-of-home will be identified (a vacant room). The time duration that the subject spends out of room will be recorded.

[0203] Aspects of the method and system disclosed herein may be used to present changes in activity routines, such as active hours, sedentary hours, or nighttime bed exit routines for each resident, using a graphical user interface (GUI), such as a GUI on a mobile app or a desktop system, as depicted in Fig. 25(b). In this instance, the system monitors and stores in memory an individual’s active and sedentary hours. Active hours indicate when the resident is walking or moving, while sedentary hours denote periods spent in bed or sitting. Additionally, the system keeps track of the frequency of bed exits overnight. Over time, the system establishes a baseline for each resident with respect to these parameters. Any degradation or changes to this baseline trigger alerts, which are displayed in the designated mobile app, as illustrated in Fig. 25(b).

[0204] Aspects of the method and system disclosed herein may be used to showcase changes in movements, such as gait patterns or the time taken to exit the bed for each resident, on a mobile app or a desktop system, as illustrated in Fig. 25(c). In this instance, the system monitors an individual’s walking speed and the time take to exit the bed. Over time, the system establishes a baseline for each resident with respect to these parameters.Any degradation or changes to this baseline trigger alerts, which are displayed in the designated mobile app, as illustrated in Fig. 25(c).

[0205] Aspects of the method and system disclosed herein can promptly identify a fall and send an alert to smartphones or desktop systems. The system may then issue an alert and / or prompt a response from a user, caregiver, or other individuals in the event of a suspected fall (Fig. 26(a)). In one aspect, the system may transmit the alert to a designated person or group of persons who are closest to the person being monitored. If there is no response in a preset time period, the system may transmit an escalation alert to the next closest additional designated person or group of persons (Fig. 26(b)). If a response is received, confirmation of the suspected fall or reporting of a false alarm may be required (Fig. 26(c)). As illustrated in Figs. 26(a) to (c), in one aspect, the alerts generated by the system may be transmitted to handheld devices (such as smartphones) running an app, whereby specifically design screens may be presented with GUIs for prompting replies and for receiving user input. In other aspects, the system and method described herein may transmit the alerts to desktop or laptop computer or the like. This confirmation or false alarm contributes to the ongoing improvement of Al models through retraining and refinement over time. For example, the fall detection system harnesses the power of machine learning to autonomously identify potential falls through the analysis of sensor data and the recognition of relevant patterns. Recognizing the inherent challenge of false positives, a companion app for caregivers has been integrated into the system. Caregivers play a pivotal role in the confirmation or dismissal of detected falls, as they receive prompt notifications through the app, enabling them to provide feedback based on their judgment. This iterative process of human feedback directly contributes to the continuous improvement of the machine learning model. By incorporating real-world insights and feedback from caregivers, the system enhances its ability to discern between actual falls and false positives, fostering a selfimproving cycle that refines the accuracy of fall detection over time. This collaborative synergy between machine learning and human input ensures a dynamic system that adapts and evolves, continually advancing its capabilities through ongoing refinement.

[0206] Aspects of the method and system disclosed herein can also be used to identify an attempt by a monitored person to exit their bed and send an alert to one or more designated persons (i.e., caregivers etc.) As above, the alerts may be transmitted to a handheld device and / or any other computer system. As illustrated in Fig. 27(a), the system sends an alert and prompts a response from the recipient of the alert in the event of asuspected bed exit attempt. In one aspect, the system may transmit the alert to a designated person or group of persons who are closest to the person being monitored. If there is no response in a preset time period, the system may transmit an escalation alert to the next closest additional designated person or group of persons (e.g., 30 seconds, as illustrated in Fig. 27 (b)). If a response is received, confirmation of the suspected bed exit attempt may be required (Fig. 27 (c)). This confirmation or false alarm contributes to the ongoing improvement of Al models through retraining and refinement over time.

[0207] In this context, the machine learning model for identifying bed exit attempts can improve itself by leveraging feedback from the alert response process. The system initiates an alert when it suspects a monitored person is attempting to exit their bed, and the designated person (caregiver, for example) receives this alert on their handheld device. The recipient’s response, either confirming or dismissing the suspected bed exit attempt, serves as valuable feedback for the machine learning model.

[0208] If the designated person confirms the bed exit attempt, it reinforces the accuracy of the model in identifying such events. On the other hand, if the alert is dismissed as a false alarm, the system gains insights into potential areas of improvement and refines its model accordingly. This feedback loop, where the machine learning model continuously learns from real-world responses, contributes to its ongoing enhancement.

[0209] The iterative nature of this feedback mechanism ensures that the model adapts to evolving patterns and nuances, becoming more precise in distinguishing actual bed exit attempts from false alarms. Over time, as more data is collected and more feedback is received, the machine learning model undergoes retraining and refinement, ultimately improving its effectiveness in identifying and alerting caregivers to bed exit attempts. This continual learning process described herein offers a unique advantage for creating a robust and adaptive system in the realm of patient monitoring and care.

[0210] Aspects of the method and system disclosed herein can identify prolonged inactivity, wandering, extended periods out of a room, prolonged absence from home, or extended in-washroom status and send alerts to designated persons as indicated above.

[0211] Aspects of the method and system disclosed herein may be used to identify the severity of a fall and determine the danger level, such as by calculating and displaying a numerical fall severity score (e.g., in percentage risk). The severity of a fall can be determined by detecting the point of contact or the specific part of the body that hits theground. For example, as illustrated in Fig. 28(a), the system has determined that an individual being monitored as fallen on their knees and hands and near their bed. Accordingly, the system may identify such fall as a moderate risk. Fig. 28(b), shows another fall example, where the individual is determined to have fallen on their hip and again near the bed. This may be categorized as a low-risk fall, as shown in the example display screen. On the other hand, Fig. 28(c) illustrates a situation where an individual has fallen in a washroom and has struck their head on the floor. In this case, the system identifies the fall as an emergency case. Thus, as will be understood, by having this type of detailed fall information available, the caregiver staff is more accurately able to triage incoming data and attend to the highest priority / highest risk situations first.

[0212] In other aspects, the method and system disclosed herein may identify the speed of a fall, which would also be indicative of the fall severity. As shown in Fig. 29 (a), a slow fall is detected from the Doppler output, while Fig. 29 (b) shows a fast fall. This information may also be factored into the severity calculation that is transmitted in the alert sent to staff, as illustrated in Figs. 28(a) to (c).

[0213] Aspects of the disclosed method and system are specifically designed to recognize and monitor body repositioning movements aimed at preventing pressure ulcers in individuals under care. This capability sophisticated analytical algorithms to continuously assess the position and movement of a resident, thereby facilitating the early detection of situations where the risk of developing pressure ulcers is increased due to prolonged immobility. By identifying periods of inactivity or inadequate repositioning, the system can alert caregivers to the need for manual intervention or automatically initiate corrective actions to alter the resident's position. The integration of Al within this framework enables the system to learn from historical data, improving its ability to predict and pre-emptively address the conditions that lead to pressure ulcer formation.

[0214] Aspects of the method and system disclosed herein are can accurately recognize and differentiate among various activities conducted in a bed or chair, including repositioning movements, attempts to exit the bed or chair, and other types of activity. This functionality is made possible through the utilization advanced data analysis algorithms, which interpret the nuances of motion and posture. By analyzing the specific characteristics of each movement, such as its duration, intensity, and pattern, the system can categorically identify the nature of the activity, whether it is a routine adjustment of position to maintain comfort and prevent health issues like pressure ulcers, or an attempt to leave the bed or chair, which couldindicate a need for assistance or potential safety risk. Additionally, the system can discern other activities that may take place in these settings, further enhancing the care and monitoring capabilities provided to individuals. The employment of machine learning and artificial intelligence technologies enables the system to continuously refine its accuracy and sensitivity, learning from each observed action to improve its predictive and analytical performance over time.

[0215] Aspects of the method and system disclosed herein can save the time of front line staff in attending to specific situations and thereby improve quality of life for residents by providing automated tracking and reporting for changes in functional status in residents.

[0216] Aspects of the method and system disclosed herein can support clinicians by prioritizing resident needs and enhancing decision making to improve efficiency and workflow.

[0217] As would be understood by persons skilled in the art, aspects of the method and system disclosed herein may be regarded as a dedicated 24 / 7 caregiver.

[0218] As various sensors become more sophisticated, additional authentication and monitoring methods may be useful to determine further information and to keep information secure in a world where hundreds of technological interactions occur for individuals every day. Biometrics are intended to provide a personal and convenient way of keeping this data secure. Aspects of the system and method disclosed herein incorporate one or more radar sensors that are adapted or configured to identify individuals from their gait pattern / activity level / vital signs / different body parts etc. or to determine further information about a user. For example, in some aspects, a high frequency radar (operating anywhere between 30 GHz to 300 GHz or other frequency bands) may be used as the sensor and data from such sensor is classified using machine learning and artificial intelligence models, which classifies individuals based on their gait pattern / activity level / vital signs / different body parts radio signatures etc. As described herein, the inventors have developed systems and methods that are able to differentiate between a large set of individuals with very high degrees of accuracy and precision. Consequently, aspects of the method and system described herein may be used as a radar-based gait pattern / activity level / vital signs / different body parts identification as an independent or an auxiliary form of two-factor authentication.

[0219] Aspects of the method and system disclosed herein can graphically display the real-time status of each resident in a large facility on a dashboard accessible via aniPad / tablet (Fig. 30 (a)) or a desktop system (Fig. 30 (b)). As can be seen, the display screen illustrates the resident’s unit number and activity. For example, with respect to Fig. 30(a), the resident in unit 101-1 is in bed and asleep; the resident in unit 103-1 is walking; the resident in unit 108-1 is not present in the unit; the resident in unit 109-1 is in the washroom; the resident in unit 113-1 is in a sitting position; and the resident in unit 107-1 has experienced a fall. It will be understood that the graphics illustrated in Figs. 30(a) and (b) are provided only by way of example. The description is not limited to any particular representation of resident information.

[0220] Aspects of the method and system disclosed herein can record and display historical data of each individual over time on a dashboard that can be displayed on a display scree, such as a screen of a handheld device, a tablet device or a computer screen. Examples of such displays are provided in Fig. 31 (a) and (b).

[0221] Aspects of the method and system disclosed herein may improve overall care using artificial intelligence that analyzes historical data autonomously and automatically.

[0222] Aspects of the method and system disclosed herein may improve overall care combining different pieces of information using people’s heath report.

[0223] Features of the Present Description

[0224] As would be understood by persons skilled in the art, various unique advantages are realized by the presently described system and method. Some of these advantages are discussed further below.

[0225] System Configuration

[0226] As noted above, one aspect of the present description is a unique radar-based systems for monitoring patients and the like. The presently described system utilizes simple radar sensors that transmit raw, unprocessed data to a server, preferably over a wired connection. As discussed above, the use of a wired connection avoids the data traffic limitations of known systems that are connected over wireless networks. It will, however, be understood that the present system can be implemented over a wireless network that does not suffer from data traffic congestion issues.

[0227] The use of simplified radar sensors and the centralized processing of data allows for the system to be easily scaled to various sizes of environments (i.e., to accommodate a large number of sensors).

[0228] As discussed above, optimal placement of radar sensors has been investigated. In one example, at least two radar sensors are placed approximately 230 cm high on adjacent walls that are perpendicularly arranged. The sensors are also angled downward, approximately 30 degrees from vertical. This orientation has been found to provide a maximum amount of coverage for a room, or other such space. It will, however, be understood that such positioning and orientation etc. may vary depending on the need.

[0229] Incorporation of Machine learning, deep learning, Generative Al

[0230] Machine learning, deep learning, Generative Al, and other methods may be used to replace all the signal processing and provide all the required information on human activities. Machine learning, deep learning, Generative Al, and other advanced computational methods stand at the forefront of a technological revolution, offering a paradigm shift in how radar data is utilized to monitor human activities. These methods present a novel approach that can entirely supplant traditional signal processing techniques, heralding a new era where complex algorithms directly interpret raw radar signals to extract meaningful insights. This direct application allows for the bypassing of conventional preprocessing steps, which are often labor-intensive and prone to introducing errors or biases. By training these models on extensive datasets that capture a vast array of human movements and activities under different conditions, they learn to recognize patterns and nuances in the data that human operators or simpler algorithms might miss. This enables the detection, classification, and analysis of human activities with unprecedented accuracy and detail. Furthermore, the adaptability and learning capability of these models mean that they continuously improve over time, refining their predictions and analyses as more data becomes available. The integration of these Al technologies into radar-based monitoring systems not only enhances their functionality but also expands their applicability, providing rich, actionable information that can be used in a wide range of applications. This approach eliminates the need for costly hardware upgrades, as the software-driven solutions can provide the required features and performance enhancements through intelligent data analysis and interpretation. As discussed above, in one aspect, the system is incorporated with an ML, deep learning model, or generative Al model for monitoring and analyzing the activities of an individual. Such activities may comprise washroom use, sleeping habits,wandering, etc., whereby the overall health of the individual can be automatically monitored and, where necessary, the appropriate personnel can be alerted and / or informed.

[0231] It will also be appreciated that the above-mentioned personnel notification process can be used to further train the model implemented by confirming the accuracy of the model’s output.

[0232] Localization Algorithms

[0233] The get_position algorithm described above provides a unique advantage by allowing the use of low resolution radar data for accurately positioning an individual. As such, the system can be easily scaled to cover many rooms in a given facility, while accurately determining the position of the individual in question.

[0234] Similarly, as also discussed above, the in-bed algorithm allows for accurately monitoring the placement of an individual when in bed.

[0235] Activity Patterns

[0236] The algorithms described herein can also serve to establish baseline activities of individuals. By comparing detected activities with such baselines, and with pre-determined thresholds for variations, the system can alert caregivers of a potential health or cognitive impairment. For example, if the system establishes a baseline wake time for an individual and the system detects repeated numbers of bed exits and non-baselines times, it may signify a health issue (such as a UTI) or some cognitive impairment that would require further attention. The system thus proactively monitors such changes in behaviour / activity and alerts the necessary personnel.

[0237] Fall Detection and Prevention

[0238] The present description incorporates predictive algorithms to accurately detect falls and distinguish them from other activities or movements. It provides real-time alerts to caregivers or relevant personnel to ensure prompt assistance. Additionally, the description allows for fall prevention measures by monitoring, for example, bed restlessness and analyzing historical data to predict and thereby minimize fall risks by providing proactive intervention. In this way, the system and method described herein allows falls, and therefore the associated harm, from being prevented.

[0239] Furthermore, unlike known wearable devices or the like that depend on kinematic information of the subject, the present system and method is able to detect both fast and slow falls, thereby allowing for detailed characterization of a fall incident.

[0240] Once a fall incident is detected, the system and method automatically informs facility staff, thereby avoiding the need for the subject to call for assistance.

[0241] Fall Severity Detection

[0242] As discussed above, the system and method described herein can, in one aspect, also assess the severity of a fall. In one example, such assessment is done by identifying the initial body part point of contact during a fall, that is, the part of the individual that first makes contact with the ground or other object (e.g., knee, elbow, head, etc.). This information may be combined with the fall speed detection mentioned above to further enhance the severity assessment. The system can then generate real-time alerts to caregivers or relevant personnel and communicate both the fall event as well as the body part point of contact to aid in severity assessment. This enables a prompt and suitable response by staff and appropriate medical attention to the subject.

[0243] This aspect of the system utilizes radar sensors to capture motion and activity data within the monitored area. The radar sensors emit signals and measure the reflected signals to gather information about the location and movement of objects. In this context, “objects” may refer to the subject or parts of the subject. Using the radar sensor data, the system applies a fall detection algorithm specifically designed for radar-based detection. The algorithm analyzes patterns and changes in the radar signals to accurately identify fall events. Upon detecting a fall, the system utilizes radar data to assess the severity of the fall. It considers various parameters such as velocity, acceleration, and displacement derived from the radar signals. Additionally, the system incorporates algorithms to determine the specific part of the body that made initial contact with the ground or other object during the fall. The system uses radar-based localization techniques to estimate the location of the fall incident within the monitored area. By analyzing the radar signals, including their time-of- flight and angle of arrival, the system can approximate the position of the fall.

[0244] For example, considering the implementation within healthcare facilities, the present system primarily targets residents who often spend significant time alone in their rooms. The significance of the present description lies in its effectiveness during these solitary instances, ensuring timely assistance and appropriate intervention. While in thepresence of other persons, severity of falls can be observed and responded to directly, reducing reliance on automated alerts or severity classification. Therefore, the present system and method are specifically designed for instances where the resident is alone, emphasizing its critical role in such scenarios. Fig. 33 shows an example flow diagram for fall classification. To detect the severity of falls, the initial step is to identify fall incidents and distinguish them from other routine activities, such as walking, jumping, transitioning from bed to walk, or bending. As shown in Fig. 33, the radar sensor is configured and for each streamed frame of raw data, a range-Doppler map is generated and stored in a buffer. Concurrently, another process operates continuously to utilize the generated range-Doppler maps for producing spectrograms. If the system detects the patient as the only occupant in the room, a fall detection algorithm is deployed. In the event of a fall incident, the severity of fall detection is then determined. The approach includes digitally simulating fall scenarios and carefully evaluating them through measurements. A deep learning model is then used to analyze the correlation between digitally simulated and physically measured data. Using the validated dataset, the machine learning model is trained to distinguish falls based on the specific body part that experiences significant impact during the event, allowing for a classification according to severity levels.

[0245] Health Monitoring and Disease Detection

[0246] By monitoring a subject’s vital signs, sleep patterns, washroom patterns, and gait / activity patterns, the present system and method captures comprehensive health data. The system then applies data analytics and Al algorithms to interpret the data and identify potential health conditions or anomalies.

[0247] In one aspect, the present system and method is adapted for analyzing a subject’s washroom behaviour, such as the duration of time spent sitting in the washroom, frequency of use etc. Prolonged periods of sitting may be indicative of constipation, which can alert caregivers to potential issues that require attention. Similarly, frequent visits to the washroom may be indicative of a UTI or other such medical condition that the subject may not be aware of or may not have reported to the care provider. Further, a subject exhibiting frequent visits to the washroom along with aimless wandering may indicate dementia. By interpreting these patterns, the system provides valuable insights into residents’ health and well-being.

[0248] In one aspect, the system correlates the above-mentioned washroom data with facility reports, such as medical records or previous health incidents relating to the subject. By combining the washroom data with such existing information, a predictive model can be developed and incorporated into the system. This predictive model utilizes machine learning algorithms and Al techniques to identify potential underlying conditions or diseases when detected behaviour is deemed to be abnormal based on the subject’s previous behaviour. Thus, by correlating washroom behavior and historical health records, the system can detect patterns and provide early indications of specific conditions and communicate same to staff and / or healthcare providers.

[0249] As will be appreciated, the system and method described herein enables proactive healthcare management by leveraging washroom data and Al-driven analysis. By identifying potential symptoms or patterns related to residents’ behaviour in the washroom, caregivers can take timely actions, such as scheduling medical evaluations, adjusting care plans, or providing targeted interventions.

[0250] Activity Recognition and Behavior Analysis

[0251] The system and method described herein utilizes Al algorithms to recognize and analyze resident activities and behaviours. It is adapted to identify specific activities such as sitting, standing, walking, or washroom usage. By monitoring activity patterns, it can detect abnormalities or changes that may indicate underlying conditions or potential risks.

[0252] In this regard, and according to one aspect, the system utilizes two radar sensors placed in the resident’s living area to capture motion and activity data. The radar sensors continuously emit and receive signals, allowing them to detect and track movements within their range. By analyzing the radar sensor data, the system applies machine learning algorithms or pattern recognition techniques to recognize and classify different activities performed by the resident. This can include activities such as walking, sitting, lying down, and other daily tasks. The system uses radar data to extract relevant features and train models to accurately recognize and categorize activities.

[0253] Using the radar sensor data, the system applies algorithms to detect periods of extended inactivity or immobility, which data may be used to reduce incidence of pressure ulcers. By analyzing the absence of, or minimal, movement patterns, the system can identify when a resident is spending a significant amount of time in bed and trigger appropriate notifications or alerts. The system also utilizes Al-based techniques, such as machinelearning or deep learning algorithms, to analyze the collected radar sensor data and make predictions about potential diseases or health conditions relating to the subject. These predictions can be used to raise alerts, provide early warnings, or prompt further medical evaluation.

[0254] Resident Identification and Personalization

[0255] The system and method allow for individual subject profiles to be created, with such profiles being based on a subject’s unique data patterns. In this way, the description allows for accurate resident identification. As discussed further below, for such identification, the system collects and interprets data related to one or more of vital signs, sleep patterns, and gait / activity patterns of the residents. This data can be acquired through various sensors and devices, such as wearable devices, bed sensors, or radar sensors, depending on the specific implementation.

[0256] The system is adapted to monitor vital signs, including parameters such as heart rate, blood pressure, respiratory rate, oxygen saturation, or any combination thereof. This information provides insights into the overall health and well-being of the residents.

[0257] The system and method also perform sleep pattern analyses. By analyzing sleep patterns, such as sleep duration, sleep stages (e.g., deep sleep, REM sleep), and sleep disturbances, the system assesses the quality of sleep of the subject. This information may then be indicative of an underlying sleep disorder or other irregularity.

[0258] The system and method also monitor the gait and activity patterns of the residents. This includes parameters such as walking speed, step count, stride length, and cadence. By analyzing these patterns, in particular by use of Al algorithms, the system can identify abnormalities or changes in mobility that may indicate underlying health issues or changes in the physical condition of the subject.

[0259] Further Enhancements

[0260] As mentioned above, the monitoring information provided by the system and method described herein may further enhanced by incorporating one or more other sensor devices. In particular, by gathering and incorporating more information concerning the subject being monitored and their surroundings, the detection accuracy of the subject system and method would be improved. Some examples of further sensors that may be incorporated into the present system and method are described below. It will be understoodthat the following examples are not intended to be limiting in any way of the scope of additional sensors that may be incorporated into the subject system and method. In particular, as would be understood, each of the additional sensors described herein provides different and discrete information that serves to complement the radar-based data discussed above. Thus, although each of the additional sensors is described individually, it will be appreciated any number or combination of such sensors may be incorporated into the system and method.

[0261] Temperature Sensors: In one aspect, the system comprises one or more temperature sensors for monitoring the ambient temperature and / or the subject’s body temperature. The data from these sensors may be monitored in real-time and / or recorded over a period of time to provide a history of body temperature and ambient temperature variations. In particular, body temperature sensors would serve to facilitate early detection of fever or hypothermia or other thermal regulation abnormalities, which may indicate potential health concerns requiring immediate attention. By integrating temperature data with radarbased activity monitoring, the system and method enhances predictive analytics for medical distress scenarios.

[0262] Door and Contact Sensors: To improve situational awareness, the system and method, in one aspect, integrates door and / or contact sensors that monitor room entry and exit activities. This integration would enable tracking of the subject’s movement patterns, unauthorized access attempts, security breaches, and the like. The data may, for example, be collected in real-time from these sensors and be processed alongside the radar inputs discussed above to provide comprehensive occupancy detection, ensuring seamless facility monitoring.

[0263] “Ground Truth Camera” for Identity Verification and People Counting: In one aspect, the system and method comprises one or more non-intrusive ground truth cameras that serve to complement the above-mentioned radar-based monitoring. As would be known to persons skilled in the art, the term “Ground truth camera” is commonly used in the machine learning, computer vision, and sensor fusion fields and refers to a reference sensor - usually a camera - that is used to capture visual data for verifying or annotating the output from another sensor (e.g., radar). In the present context, one or more ground truth cameras assists in the identification of subjects / individuals, counting people, and activity verification. Thus, a "ground truth camera" as used herein refers to a reference imaging device that is used to validate and refine the data captured by other sensors of the presentsystem and method, such as the radar sensors. In the present context, such camera(s) provides a non-intrusive visual verification tool for validating the identity and number of individuals detected by the present radar system. The incorporation of data obtained from such camera(s), reduces the incidences of false positives. In particular, the data acquired from the ground truth camera would serve to refine the classification of detected movements and ensure higher accuracy in distinguishing between multiple individuals within the monitored environment. Importantly, all camera data would be transmitted only to the internal radar sensor for processing, with no external connections, wireless transmissions, or external network access thus ensuring that the collected visual data is maintained private and secure within the local facility.

[0264] Two-Way Al-Based Voice Communication: In another aspect, the system and method includes a voice communication equipment to allow voice interactions with a subject. In a preferred aspect, the system and method includes an Al-driven two-way voice communication module, leveraging advanced Al models, such as known generative Al and natural language processing, to enable direct interaction with monitored subjects. This module could allow caregivers, security personnel, or emergency responders to engage with individuals in real time, facilitating proactive interventions. The Al component is adapted to process and generate natural language interactions, providing automated guidance, dynamic conversation capabilities, and emergency assistance while minimizing unnecessary intrusions into privacy.

[0265] Integration into Building Automation Systems: To further enhance usability and system-wide connectivity, the monitoring system and method described is, in another aspect, configured to interface with building automation frameworks, including but not limited to smart home, lighting, and security infrastructures. Such integration allows for automated control of environmental parameters based on detected activities, improving occupant safety and energy efficiency. For instance, the system and method could trigger emergency lighting upon detecting a fall or unlock doors when medical distress is detected. Additionally, the system and method could proactively adjust the environment based on predicted activity. For example, if the system and method detects that a resident is beginning to get out of bed during nighttime hours, it could automatically turn on soft ambient lighting to assist in preventing falls. Similarly, it could adjust room temperature or activate safety mechanisms in response to changes in occupant behavior.

[0266] Hierarchical Device Network (Centralized-Distributed Model): In another aspect, the system and method is comprised within a structured centralized-distributed architecture to optimize wired sensor deployments. In this configuration, a central radar sensor with a high-performance processing unit aggregates and processes data from multiple distributed radar sensors. As would be appreciated by persons skilled in the art, this approach eliminates the need, for example, to send individual sensor data to an edge server within the facility, thereby enhancing efficiency and reducing latency. The central processing radar sensor is preferably adapted to perform local computations, thus improving system responsiveness while ensuring data privacy within the facility. Furthermore, this aspect of the present system and method reduces the need for extensive cabling, as the distributed sensors would rely on the central unit for processing, minimizing infrastructure complexity and installation costs.

[0267] Gesture Recognition for High-Security Applications: In another aspect, the system and method includes a gesture recognition module. Such module may be tailored for example, for high-security environments, such as correctional facilities and psychiatric institutions. In this aspect, the system and method is adapted to analyze subtle and / or rapid hand and body movements detected by the radar sensors, which may be indicative of aggression, self-harm, or unauthorized interactions. By detecting these gestures, such as in real time, the system and method could enable preemptive intervention, reducing the likelihood of harmful incidents. It will be understood that this module will process radar data indicative of such gestures either independently or in combination with other visual data acquired, for example, from camera systems such as discussed above.

[0268] Care Planning and Predictive Analytics: In one aspect, the system and method forms a data acquisition component for planning a personalized care support program by leveraging Al-driven predictive analytics. For example, by monitoring an individual’s behavior, such as changes in movement patterns and other physiological signals, the system and method may be adapted to identify trends and predict potential health risks in a subject. This capability would allow caregivers and healthcare providers to create customized care plans based on real-time and historical data. Furthermore, the system and method may be adapted to automatically generate individualized care plans by analyzing activity patterns and health indicators. For instance, if the system detects a progressive decline in a subject’s mobility, it may suggest adjustments in physical therapy routines or other preventive interventions. Additionally, the system and method is configured to provideautomated reports and alerts to caregivers, ensuring proactive and timely care adjustments while dynamically updating recommendations based on ongoing monitoring.

[0269] Health Risk Prediction Based on Historical Data: In another aspect, the system and method described herein is adapted to utilize historical data from radar and other integrated sensors to identify deviations from normal patterns and identify potential health concerns before they escalate. For example, a sudden decrease in mobility or irregular movement patterns could indicate that the subject may be at an increased risk of falls, thereby prompting proactive and preventive intervention. Additionally, changes in bathroom usage frequency or activity levels could signal potential urinary tract infections (UTIs) or other underlying health conditions. In yet another aspect, the system and method described herein are adapted to analyze long-term behavioral trends to detect early signs of cognitive decline, cardiovascular issues, or other chronic conditions. By monitoring and analyzing these data points, the system and method provides early alerts and recommendations to caregivers, ensuring timely medical attention and reducing the risk of complications. In a further aspect, the present system and method is adapted to provide automated reports and alerts to caregivers, ensuring proactive and timely care adjustments while dynamically updating recommendations based on ongoing monitoring.

[0270] Bed Sore Prevention and Repositioning Alerts: In a further aspect, the presently described system and method is adapted to monitor prolonged inactivity and lack of movement in bedridden individuals thus serving to detect the possibility of bed sores. For example, by analyzing movement patterns, the system and method detects when a subject has remained in the same position for an extended period of time and assesses the risk of pressure ulcers. This analysis may be performed with a predictive algorithm, for example. If a risk threshold is exceeded, the system and method could initiate an automated alert to caregivers, prompting them to assist in repositioning the individual. Additionally, in a further aspect, the system and method is adapted to issue voice prompts directly to the subject, encouraging self-repositioning when possible. For this, the system incorporates one or more speakers for broadcasting messages to the location of the subject. Thus, by implementing proactive movement reminders and automated tracking, the system and method enhances the subject’s comfort, reduces medical complications, and improves the overall care quality for individuals with limited mobility.

[0271] Expanded Activity Detection for Risk Assessment and Intervention

[0272] To further improve safety and response mechanisms, another aspect of the system and method described herein incorporates advanced behavioral analytics for detecting destructive behaviors, self-harm tendencies, and medical distress. Some of these features are described above. The following enhancements are incorporated into the system and method to refine its ability to identify and respond to critical incidents:

[0273] Destructive Behavior Detection: As noted above, in one aspect, the system and method is adapted to identify violent or aggressive actions by subjects being monitored. Such actions may include physical altercations with other individuals, the striking of fixtures and / objects, such as walls and doors, and destructive interactions with the subject’s environment. In one aspect, the system and method described herein incorporate analysis algorithms for detecting fire-setting attempts, forced entry behaviors, and aggressive reaching, tripping, or grabbing of others. These detections could trigger real-time alerts to security personnel or caregivers, ensuring immediate intervention and risk mitigation.

[0274] Self-Harm Behavior Monitoring: In one aspect, the present radar-based system and method includes classifiers, or classification algorithms, for detecting self-harm tendencies, such as head-banging against walls or doors, self-inflicted injuries using sharp objects, and ingestion of non-edible substances (e.g., paint, caulking materials). In one aspect, the system is adapted to identify high-risk behaviors such as strangulation / hanging attempts and drowning risks in confined spaces such as toilets or sinks. As above, upon detecting potential self-harm incidents, the system is adapted to initiate automated emergency alerts and / or activate voice communication features to de-escalate the situation.

[0275] Medical Distress Recognition: In one aspect, the present system and method features real-time physiological monitoring, including but not limited to monitoring of heart rate, respiratory rate, and / or temperature. In this way, the system and method serves to detect potential life-threatening medical conditions of the subject being monitored. Similarly, in another aspect, the system and method are adapted to detect fluctuations in a subject’s heart rate (such as when a subject is experiencing tachycardia or bradycardia) and respiratory distress signals (e.g., apnea, hyperventilation). In one aspect, the system and method are adapted to automatically flag the abnormal status and alert the necessary person(s) thus prompting emergency response actions. In another aspect, the system and method are adapted to monitor body temperature irregularities to identify fever or hypothermia. As discussed above, such monitoring is accomplished using one or more temperature sensors incorporated into the system and method. Such monitoring thusensures timely medical intervention. Additionally, in yet a further aspect, the system and method are adapted to monitor a subject’s perspiration. As would be understood, such monitoring allows the system and method to identify underlying conditions such as diabetes, heart failure, anxiety disorders, or hyperthyroidism.

[0276] As would be understood, by integrating the one or more of the above-mentioned enhancements, the present system and method provides a holistic, non-intrusive, and privacy-conscious monitoring solution that is applicable across healthcare, eldercare, security, and smart facility management domains. These advancements thus significantly improve the ability to predict, detect, and respond to critical incidents, ensuring a safer and more responsive monitoring ecosystem.

[0277] Although the above description includes reference to certain specific aspects, various modifications thereof will be apparent to those skilled in the art. Any examples provided herein are included solely for the purpose of illustration and are not intended to be limiting in any way. Any drawings provided herein are solely for the purpose of illustrating various aspects of the description and are not intended to be drawn to scale or to be limiting in any way. The scope of the claims appended hereto should not be limited by the preferred aspects set forth in the above description but should be given the broadest interpretation consistent with the present specification as a whole. The disclosures of all references in the present description herein are incorporated herein by reference in their entirety.

Claims

WE CLAIM:1 . A system for monitoring a subject in a three dimensional environment comprising one or more rooms, the system comprising:- one or more radar sensors, each of the radar sensors comprising at least one radar wave transmitter for transmitting radar waves and at least one radar wave receiver for receiving reflections of the transmitted radar waves;- a processor configured to receive data from the one or more radar sensors, the data being indicative of the reflected radar signals received by the at least one radar wave receiver;- a memory in communication with the processor, the memory having stored therein one or more executable algorithms for processing the data received from the one or more radar sensors;- wherein the processor is configured to monitor the movement of the subject in the environment in three dimensions based on the data received from the one or more radar sensors.

2. The system of claim 1 , wherein the one or more radar sensors are millimeter wave radar sensors.

3. The system of claim 1 or 2, wherein the one or more radar sensors generate radar waves at a frequency of about 1 GHz to about 300 GHz.

4. The system of any one of claims 1 to 3, wherein the environment comprises multiple rooms, and wherein each room is provided with at least one of the radar sensors.

5. The system of any one of claims 1 to 4, wherein one of the executable algorithms is a presence-absence detection algorithm, and wherein the processor is configured to execute the presence-absence detection algorithm to detect presence or absence of a subject and to detect entry and exit of the subject from a first room based on the data received from the one or more radar sensors.

6. The system of claim 5, wherein the processor is further configured to analyze the entry and exit activities of the subject to and from the first room over a period of time and to record the entry and exit activities in the memory.

7. The system of claim 5 or 6, wherein the processor is further configured to determine the presence of the subject in the room.

8. The system of claim 7, wherein the processor is configured to detect and analyze the breathing of the subject for determining the presence of the subject in the room.

9. The system of claim 8, wherein the processor is configured to detect and analyze micro-Doppler signatures generated by chest movements of the subject for detecting and analyzing the breathing of the subject.

10. The system of claim 9, wherein the presence-absence detection algorithm incorporates machine learning and / or deep learning models for analyzing the micro-Doppler signatures.11 . The system of any one of claims 5 to 10, wherein the processor is further configured to generate an alert of the entry and exit activities of the subject and to transmit the alert to a user device.

12. The system of claim 11 , wherein the processor is configured to display the alert on a graphical user interface on the user device.

13. The system of any one of claims 1 to 4, wherein one of the executable algorithms is a fall severity detection algorithm, and wherein the processor is configured to execute the fall severity detection algorithm and to detect and analyze a fall of a subject based on the data received from the one or more radar sensors.

14. The system of claim 13, wherein the fall severity detection algorithm analyzes the data received from the one or more radar sensors to detect abrupt changes in speed of movement of the subject and, responsive to detecting an abrupt change in speed, to generate a signal that is indicative of a fall incident.

15. The system of claim 14, wherein the fall severity detection algorithm further analyzes the data received from the one or more radar sensors to detect a body part of the subject to contact the ground in the event of a fall incident.

16. The system of 14 or 15, wherein the process is configured to detect the abrupt changes in speed of movement of the subject by comparing movements of the subject to at least one baseline movement speed.

17. The system of any one of claims 14 to 16, wherein the fall severity detection algorithm further incorporates results from a machine or deep learning model to further analyze the data from the one or more radar sensors to characterize the movement of the subject during the fall incident and to thereby confirm the indication of a fall.

18. The system of any one of claims 14 to 17, wherein the fall severity detection algorithm further detects the subject’s position in the three dimensional environment to determine if a vertical change in position has occurred to confirm the indication of a fall incident.

19. The system of any one of claims 14 to 18, wherein the fall severity detection algorithm further detects the location of the subject and compares such location with detected fall incident and confirms the fall incident when the detected location of the subject is the location of the detected fall incident.

20. The system of any one of claims 14 to 19, wherein the processor is further configured to generate an alert of the fall incident and to transmit the alert to a user device.21 . The system of claim 20, wherein the processor is configured to display the alert on a graphical user interface on the user device.

22. A method for monitoring a subject in a three dimensional environment comprising one or more rooms, the method comprising:- providing one or more radar sensors, each of the radar sensors comprising at least one radar wave transmitter for transmitting a radar signal and at least one radar wave receiver for receiving a reflection of the radar signal;- transmitting the reflected radar signals to a processor;- processing the reflected radar signals with a presence-absence detection algorithm to determine entry and exit activities of the subject from at least a first room of the environment.

23. The method of claim 22, further comprising storing the entry and exit activities of the subject in a memory.

24. The method of claim 22 or 23 further comprising processing the reflected radar signals to detect and analyze the breathing of the subject to determine the presence of the subject in the first room.

25. The method of claim 24, wherein the breathing of the subject is detected and analyzed using micro-Doppler signatures generated by chest movements of the subject.

26. The method of claim 25, wherein the presence-absence detection algorithm incorporates machine learning and / or deep learning models for analyzing the micro-Doppler signatures.

27. The method of any one of claims 22 to 26, further comprising generating an alert of the entry and exit activities of the subject and transmitting the alert to a user device.

28. The method of claim 27, wherein the alert is displayed on a graphical user interface on the user device.

29. A method for monitoring a subject in a three dimensional environment comprising one or more rooms, the method comprising:- providing one or more radar sensors, each of the radar sensors comprising at least one radar wave transmitter for transmitting a radar signal and at least one radar wave receiver for receiving a reflection of the radar signal;- transmitting the reflected radar signals to a processor;- processing the reflected radar signals with a fall severity detection algorithm to detect and analyze a fall incident of a subject.

30. The method of claim 29, wherein the method comprises detecting abrupt changes in speed of movement of the subject and, in response to detecting an abrupt change in speed, generating a signal that is indicative of a fall incident.31 . The method of claim 29 or 30, further comprising detecting a body part of the subject to contact the ground in the event of a fall incident.

32. The method of any one of claims 29 to 31 , wherein the detection of abrupt changes in speed of movement of the subject comprises comparing the detected speed of movement of the subject to at least one baseline movement speed.

33. The method of any one of claims 29 to 32, wherein the detection and analysis of the data from the one or more radar sensors is processed with a machine or deep learning model to characterize the movement of the subject during the fall incident and to thereby confirm the indication of a fall.

34. The method of any one of claims 29 to 33, wherein the data from the one or more radar sensors is processed to determine the subject’s position in the three dimensional environment and to determine if a vertical change in position has occurred to confirm the indication of a fall incident.

35. The method of any one of claims 29 to 34, further comprising processing the data from the one or more radar sensors to determine the location of the subject and wherein the location of the subject is compared to the location of the detected fall incident, and confirming the fall incident when the detected location of the subject is the location of the detected fall incident.

36. The method of any one of claims 29 to 35 further comprising generating an alert of the fall incident of the subject and transmitting the alert to a user device.

37. The method of claim 36, wherein the alert is displayed on a graphical user interface on the user device.

Images