Radar-based human presence monitoring system with edge filtering and cloud-based feedback

Abstract

A human presence monitoring system includes a radar-based sensing device configured for installation in a residential environment and a cloud-based server configured to receive and analyze sensing data. The device emits frequency-modulated continuous-wave radar signals and performs on-device signal processing using distance, energy, and motion gating to detect and classify human presence. A visual indicator provides real-time feedback to the user. The device aggregates radar data into a compressed two-dimensional histogram and transmits periodic reports via a cellular or Wi-Fi connection. The server applies machine learning classifiers trained on historical activity and client-specific circadian patterns to assess reported data and determine alert conditions. Updated classification parameters may be transmitted back to the device without user intervention, allowing the system to adapt over time. A companion application enables authorized contacts to receive daily reports, view activity history, configure alert preferences, and manage support circle communications.

Description

RADAR-BASED HUMAN PRESENCE MONITORING SYSTEM WITH EDGEFILTERING AND CLOUD-BASED FEEDBACKRELATED APPLICATION

[0001] This application claims priority benefit of U.S. Provisional Patent Application No. 63 / 675,249 filed July 24, 2024, which is hereby incorporated by reference in its entirety.TECHNICAL FIELD

[0002] The disclosure relates generally to human presence detection systems, and more particularly to radar-based monitoring devices for residential use that perform local signal processing and communicate with remote servers for activity classification, adaptive threshold adjustment, and alert generation. The system is particularly suited for supporting the safety and autonomy of independently living older adults.BACKGROUND INFORMATION

[0003] A significant portion of older adults in the United States live alone. According to the Pew Research Center, 27% of adults aged 60 and older, or around 16.7 million people, lived alone in 2020, and this population is projected to continue rising. These “solo agers” express a strong preference for continuing to live independently as long as possible. However, living alone does place them at risk in case of a sudden illness, injury, or death, as well as environmental threats due to storms, extreme heat, and power outages.

[0004] Certain prior art systems, including those described in International Patent Application Publication No. WO2021 / 118570 and U.S. Patent No. 8,742,935, have employed radar-based sensing for detecting falls or monitoring human presence in indoor environments. These systems typically rely on continuous-wave radar and centralized classification techniques to identify motion patterns or anomalies.

[0005] U.S. Patent Application Publication Nos. 2023 / 0263393 and 2022 / 0230746 describe multi-sensor monitoring systems that incorporate environmental and activity data to infer user behavior or risk states. While these approaches may be suitable for certain applications, they often involve complex installations, require constant network connectivity, or depend on high-bandwidth data transfers. These and other limitations in existing technologies leave open the need for more robust, adaptable, and user-centered presence monitoring solutions in residential settings.SUMMARY OF THE DISCLOSURE

[0006] A human presence monitoring system is disclosed, designed to support the safety and autonomy of independently living older adults through unobtrusive sensing and intelligent reporting. The system includes a radar-based monitoring device configured to operate in a residential environment, either plugged directly into a wall outlet or positioned on a stand and powered via USB. The device includes a millimeter- wave frequency- modulated continuous-wave (CWFM) radar sensor and performs real-time on-device signal processing using configurable distance, energy, and motion gating to detect likely human presence while filtering out non-human movement such as that from pets or environmental noise.

[0007] The monitoring device aggregates radar returns into a two-dimensional histogram of energy and distance values, compresses this data, and transmits periodic reports over a cellular or Wi-Fi connection to a remote server. A visual indicator on the device provides immediate feedback to the user, such as a green light to indicate human presence and active system connectivity, or a red light to indicate communication loss.

[0008] A cloud-based server receives and stores periodic reports, applies machine learning models trained on historical data and client-specific circadian rhythms, and determines whether alert conditions exist. The server may automatically transmit updated classification parameters to the monitoring device, thereby adjusting real-time gating behavior without requiring user intervention.

[0009] An optional companion application enables authorized contact individuals to receive automated daily reports or “up and around” notifications, review historical activity charts, manage recipient preferences, and view client well-being status. The application also supports administrative functions for configuring alert settings, tagging ambiguous activity data for supervised classifier training, and refining overall system behavior through human- in-the-loop machine learning workflows.

[0010] This integrated system balances simplicity, privacy, and adaptability to deliver reassurance and actionable insights while preserving the dignity and independence of the monitored individual.

[0011] Additional aspects and advantages will become apparent from the following detailed description of various embodiments, taken in conjunction with the accompanying drawings.BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWINGS

[0012] To easily identify the discussion of any particular element or act, the most significant digit or digits in a reference number refer to the figure number in which that element is first introduced.

[0013] FIG. l is a block diagram illustrating components of a human presence monitoring system for use in a residential environment, including a cloud-based server, a monitoring device located in a client’s residence, wireless communication connections, and interfaces with contact individuals via alert messages and a cross-platform application.

[0014] FIG. 2 is a perspective view of a monitoring device in a plug-mount configuration, showing a “night light” enclosure mounted in a wall outlet.

[0015] FIG. 3 is a perspective view showing internal components of the monitoring device in the “night light” configuration, including the radar module, microprocessor, wireless communication components, and environmental sensors.

[0016] FIG. 4 is a perspective view showing a monitoring device in a countertop configuration with an external stand and USB charger.

[0017] FIG. 5 is a rear perspective view of the countertop configuration showing the USB connector, mounting features, and optional keyhole notch for wall hanging.

[0018] FIG. 6 is a flowchart showing signal processing steps performed by the system, including real-time, on-device classification and cloud-based decision logic for activity recognition and alert generation.

[0019] FIG. 7 is a flowchart showing supervised functions used to refine device behavior and update real-time gating thresholds.

[0020] FIG. 8 is a histogram chart with gating thresholds for real-time on-device decisions.

[0021] FIG. 9 is a flowchart of a process for extracting circadian features, such as wake-up time, bedtime, and sleep continuity, detecting sentinel trends and notifying well-being partners of concerning trends.

[0022] FIG. 10 is a flowchart illustrating a server-based process for classifying human presence using radar echo histograms and template-based activity profile matching.

[0023] FIG. 11 is is a flowchart illustrating an override and adaptation process in which server-based classification results may override on-device decisions, followed by retrospective evaluation and potential retraining of classification parameters.

[0024] FIG. 12 is a diagram showing a text message containing a daily status report and a link to a well-being update landing page.

[0025] FIG. 13 is a user interface screenshot showing a well-being update page reached via a text message link, including alert status, recent activity, and access to historical data.

[0026] FIG. 14 is a chart view showing daily activity and temperature data over multiple days, using graphical bars to indicate hourly activity and report coverage.

[0027] FIG. 15 is a list view showing an hourly log of client activity and temperature readings, including active hours, periods of inactivity, and a midnight separator.

[0028] FIG. 16 is a user interface screen showing an advanced sensor analysis display with a radar histogram, classifier settings, and tagging tools for machine learning updates.

[0029] FIG. 17 is a graphical histogram showing radar data collected during a time interval with no human activity, with echoes confined to low-energy bins.

[0030] FIG. 18 is a graphical histogram showing radar data indicative of human activity, with echoes distributed across medium range and higher energy bins.

[0031] FIG. 19 is a graphical histogram showing radar data indicative of pet activity, with localized, short-range echoes consistent with a small animal.

[0032] FIG. 20 is a user interface screen showing a contact management interface within a companion application, used to configure and connect with members of the client’s support circle.

[0033] FIG. 21 is a user interface screen showing a well-being summary page within the companion application, including presence sensor status, last activity, and temperature data.

[0034] FIG. 22 is a user interface screen showing a setup interface for configuring daily report message recipients, delivery methods, and delivery conditions.

[0035] FIG. 23 is a user interface screen showing a setup interface for configuring “Up and Around” notifications, including recipient selection and delivery rules.

[0036] FIG. 24 is a flowchart of a process for monitoring human presence in a residential environment, in accordance with one embodiment.

[0037] FIG. 25 is a block diagram illustrating hardware resources that may be used to implement components of the monitoring device or server functions, including processors, memory, communication resources, peripheral interfaces, and machine-readable instructions.DETAILED DESCRIPTION OF EMBODIMENTS

[0038] A powerful tension between the desire for autonomy and the need for physical safety creates a significant challenge in need of a practical solution that supports the solo ager’s independence and autonomy while providing reassurance that someone will know ifthey are incapacitated. Optimizing this communications link can also offer a profound emotional benefit by improving the relationship between the solo ager and their family and community.

[0039] Existing solutions for monitoring the safety and well-being of independently living older adults suffer from a range of limitations that reduce their effectiveness or acceptability. Traditional home security systems, while useful for detecting fires or intrusions, are not configured to detect more common risks such as medical emergencies or environmental hazards. These systems often rely on passive infrared (PIR) motion sensors, which can be triggered falsely by sunlight, heat sources, or pets. Installation typically requires an in-person service call, and operation may be burdensome for older adults, particularly when remembering to arm or disarm the system. False alarms may also result in fees and cause unnecessary distress or embarrassment.

[0040] Some commercially available systems incorporate cameras or microphones to monitor activity, but these raise significant privacy concerns. While such systems may be acceptable in institutional or advanced care settings, they are generally unsuitable for independent-living adults who value autonomy and dignity. These systems also depend on high-speed internet connectivity and may be susceptible to hacking, further discouraging adoption among older users.

[0041] Wearable medical alert buttons are another common solution but are frequently perceived as stigmatizing or inconvenient. Many older adults discontinue their use shortly after adoption, leaving the devices in a drawer or on a nightstand. Wearable versions require battery charging or replacement, which may be difficult for individuals with dexterity or memory challenges. Nightstand-based versions may be unreachable in the event of a fall or other incapacitating event.

[0042] Smartwatches offer additional features such as emergency call capabilities, fall detection, and physiological monitoring. However, they share many of the same limitations as call buttons, including the need for frequent charging and reliance on small screens or companion smartphone apps that may be difficult to operate. Their cost (both initial and ongoing) may also place them out of reach for older adults living on fixed incomes.

[0043] These limitations highlight the need for a more intuitive, non-stigmatizing, and technically robust solution that can monitor activity reliably while respecting the preferences and constraints of older adults who wish to live independently.

[0044] FIG. 1 shows an example human presence monitoring system 100. Human presence monitoring system 100 includes a secure database 102 and server functions 104 coupled tosecure database 102. Secure database 102 and server functions 104 operate in the cloud and are accessible, with appropriate security and encryption, via the Internet. A cross-platform application operable on a mobile device 106 or computer web browser 108 communicates with server functions 104 and secure database 102 via the Internet connection of mobile device 106 or computer web browser 108. A single instance of secure database 102 and associated server functions 104 may be scaled to support multiple clients using human presence monitoring system 100.

[0045] Human presence monitoring system 100 further includes a radar-based monitoring device 110 located in a client’s residence. Radar-based monitoring device 110 is in its countertop configuration (shown in greater detail in FIG. 4 and FIG. 5) and includes a radar sensor configured to emit frequency-modulated continuous-wave radar signals and receive reflections within a field of view encompassing the client’s normal daily activities. In some embodiments, radar-based monitoring device 110 is configured as a common household object, such as a night light or picture frame. For a night light configuration, radar-based monitoring device 110 may be mounted directly into a hallway electrical outlet. For a picture frame configuration, radar-based monitoring device 110 may be placed on a countertop, such as in a kitchen, and powered via a USB cable connected to a power supply.

[0046] Radar-based monitoring device 110 includes wireless communication circuitry configured to transmit data via a Wi-Fi connection through a router 112 or via a cellular connection through a nearby cellular access point 114. In either case, data is transmitted to server functions 104 and stored in secure database 102. Radar-based monitoring device 110 transmits periodic reports, such as hourly summaries, in a compact format including classification and histogram-based radar sensing data.

[0047] In some implementations, human presence monitoring system 100 may further include one or more auxiliary sensors (not shown) located within the client’s residence. The auxiliary sensors may include motion sensors, proximity sensors, environmental sensors, or biometric sensors, and may communicate wirelessly with radar-based monitoring device 110 using Bluetooth Low Energy, Zigbee, or another low-power wireless protocol. Data received from auxiliary sensors is transmitted by radar-based monitoring device 110 to server functions 104 for analysis alongside data from radar-based monitoring device 110.

[0048] Server functions 104 include a pattern analyzer that is triggered when new data is received. The pattern analyzer compares incoming activity, behavioral, physiological, and environmental data against historical or expected patterns. Based on deviations from expected patterns, the pattern analyzer may detect an alert condition and generate one ormore alert messages. Alert messages may be delivered via email 116 or SMS 118 to designated contact individuals. A well-being update 120 may be embedded in each message, linking to a landing page 122 accessible without login credentials.

[0049] Alert trigger conditions may be set by default, specified by client preferences, or generated via machine learning. For example, alert conditions may include absence of activity by a certain time, abnormal temperature combined with inactivity, or nighttime activity inconsistent with learned sleep patterns. In some embodiments, alerts may be triggered if no data is received for a defined interval, indicating a power or network outage.

[0050] When alert conditions are detected, server functions 104 may generate escalating alerts delivered to contact individuals in a predefined sequence. Contact individuals may include neighbors, friends, family, or community organizations. Although public safety agencies are not contacted directly by human presence monitoring system 100, contact individuals may independently escalate concerns if they determine the client is at risk.

[0051] Turning now to FIG. 2, radar-based monitoring device 110 is shown in a plugmount configuration 200, also referred to here as a “night light” configuration. This configuration presents a familiar and unobtrusive appearance, enabling integration into typical residential environments and supporting passive monitoring with no client interaction.

[0052] Radar-based monitoring device 110 includes an enclosure assembly 202 comprising an enclosure back 204 and an enclosure front 206. Enclosure back 204 and enclosure front 206 are fastened together to form a unitary housing shaped and styled to resemble a conventional night light.

[0053] Enclosure assembly 202 is mounted directly into a wall outlet 208, which supplies electrical power for operation of radar-based monitoring device 110. In some embodiments, wall outlet 208 may be located in a hallway or other area where the client’s routine movement places them within the radar field-of-view of radar-based monitoring device 110.

[0054] Enclosure front 206 includes a space for label 210, which may accommodate a visual identifier, branding element, or other form of personalization. An LED glow area 212 is also provided on enclosure front 206 and is configured to offer visual feedback to the client. In various embodiments, LED glow area 212 may illuminate in different colors to indicate operational status, such as detection of human presence or confirmation of end-to- end system connectivity with server functions 104 of FIG. 1.

[0055] Turning now to FIG. 3, radar-based monitoring device 110 is shown with its enclosure front 206 removed to illustrate internal components in the “night light”configuration. Enclosure front 206 is normally fastened to enclosure back 204 by means of one or more dovetail clips 302 at the lower corners and one or more countersunk screw bosses 304 at the upper corners.

[0056] A printed circuit board assembly 306 is mounted within the enclosure and contains the core electronics of monitoring device 110. A USB plug 308 on printed circuit board assembly 306 engages with a removeable USB charger 310, which fits into a cavity in enclosure assembly 202 and draws power from wall outlet 208 (FIG. 2). Printed circuit board assembly 306 includes a microprocessor module 312 that coordinates device operation and data handling.

[0057] A Wi-Fi and Bluetooth (2.4 GHz) antenna 314 provides wireless connectivity for in-home communication, and a cellular modem module 316 is optionally included for communication in client environments where Wi-Fi connectivity is unavailable. A cellular antenna 318 (not visible in FIG. 3) is coupled to cellular modem module 316 to facilitate transmission over cellular networks.

[0058] A millimeter-wave continuous-wave frequency-modulated (CWFM) radar module 320 is mounted to printed circuit board assembly 306 and is configured to detect motion and presence within a specified sensing field. Indicator RGB LED 322 is positioned near the upper edge of the device and provides visual output to the user, for example by displaying green or red status illumination to indicate system state or connectivity.

[0059] A temperature / humidity sensor 324 is located in a lower region of the enclosure and is thermally isolated from heat-generating components of printed circuit board assembly 306. Ventilation slots 326 are located adjacent to temperature / humidity sensor 324 to admit ambient air for environmental sensing. Additional ventilation slots 328 are placed at the top and 330 at the side of the enclosure to dissipate heat generated by internal components during operation.

[0060] Not explicitly shown in FIG. 3 but included in some embodiments, a rechargeable backup battery may be positioned behind printed circuit board assembly 306 to maintain limited operation in the event of a power failure. Additional integrated circuits (not separately labeled) may be included on printed circuit board assembly 306 for functions such as power management and orientation sensing.

[0061] With reference to FIG. 4, radar-based monitoring device 110 is shown in a countertop configuration 400. In this embodiment, radar-based monitoring device 110 is positioned vertically and supported by a removable countertop stand 402. Removable countertop stand 402 is designed to provide a stable and slightly angled platform formonitoring device 110, enabling radar sensing directed into the room from a raised surface such as a kitchen counter. This configuration is advantageous in scenarios where wall outlet mounting is impractical or where specific radar positioning is desired to reduce interference from pets or furniture.

[0062] Like in its plug-mount configuration 200, monitoring device 110 in its countertop configuration 400 is powered by a removeable USB charger 310, which is plugged into a wall outlet and connected to the device via a male-to-female USB extension cable.

[0063] In FIG. 5, a rear view of countertop configuration 400 is shown. A cavity 502 is formed in the rear of enclosure assembly 202 to accommodate removeable USB charger 310 that is slidably removable from cavity 502. A USB male plug 504 is positioned within cavity 502 and is electrically coupled to internal printed circuit board assembly 306. USB male plug 504 engages with a USB female connector (not shown) to provide electrical power to radar-based monitoring device 110 when in use.

[0064] Also visible in FIG. 5 is a keyhole notch 506 located on the rear surface of enclosure assembly 202. Keyhole notch 506 allows radar-based monitoring device 110 to be mounted on a vertical surface, such as a wall, using a screw or nail, in a wall-mount configuration. In this configuration, removable countertop stand 402 may be removed, folded flat, or otherwise repositioned to enable flush wall mounting.

[0065] In light of plug-mount, countertop, and wall-mount configurations, skilled persons will appreciate that radar-based monitoring device 110 includes a multi-purpose design that allows for flexible deployment in a variety of residential layouts.

[0066] Turning now to FIG. 6, a signal processing process 600 is shown as a flowchart, illustrating both real-time, on-device functions and server database and functions in human presence monitoring system 100. The flowchart is divided vertically into two operational domains: real-time, on-device functions (left side) and server database and cloud functions (right side), with data exchanged between them over a cellular or Wi-Fi network connection 602.

[0067] On the device side, signal processing begins at radar data acquisition 604, where a radar data stream is received comprising energy, distance, and motion parameters sampled approximately ten times per second. Radar-based monitoring device 110 accumulates 606 these data into a two-dimensional histogram indexed by distance and signal energy.

[0068] Process 600 simultaneously makes a gating determination 608, in which the data stream is evaluated using configurable gating logic, including energy, distance, and motion thresholds. For example, an energy threshold may be set to accept radar echoes fromhuman-sized objects while rejecting those from dogs and cats. A distance threshold may be set to accept radar echoes within a specific room while rejecting those from more distant rooms. A motion threshold may be set to accept only movement velocities typical of human activities while rejecting slower and faster velocities. Note that even though the distance and energy appear widely dispersed (see, e.g., histogram in FIG. 18), the on-device distance gating operates against the raw data stream, not the accumulated hourly histogram. In the raw data stream, successive samples carry motion information by subtracting the previous sample’s distance from the current sample’s distance, and dividing by the time interval between samples.

[0069] These thresholds may be set locally or received as updated parameters from the server-side classifier shown and described later with reference to FIG. 7. If the gating criteria are not satisfied, processing returns to radar data acquisition 604 to continue data acquisition. If the criteria are met, processing proceeds to FIFO buffer 610.

[0070] Next, radar-based monitoring device 110 performs a count threshold check 612 to evaluate whether a sufficient number of successive events in FIFO buffer 610 pass a count threshold. If the threshold is not met, process 600 loops back to radar data acquisition 604. If the threshold is satisfied, process 600 proceeds to activity detection 614 to declare a local activity event and activate a local indicator (e.g., green LED glow area 212).

[0071] At event count per hour 616, radar-based monitoring device 110 determines whether the number of activity events in the current hour meets a predefined hourly gating threshold. If so, radar-based monitoring device 110 declares the client active 618 for the hour.

[0072] Radar-based monitoring device 110 then uploads 620 a report to its server database 622 including hourly activity status, compressed histogram data, environmental sensor readings, and other metadata. This report is transmitted periodically, such as once per hour.

[0073] On the server side, a sentinel process 624 is initiated, triggered on an hourly basis to evaluate incoming device data in database 622. A pattern classifier 626 is then applied to analyze the hourly histogram, and if necessary update the activity decision (activity detection 614), which was made by device 110 in real-time. The hourly data is tagged if pattern classifier 626 overrules the real-time device decision, or if the pattern is an outlier that produces a low-confidence decision from pattern classifier 626.

[0074] Unlike the gating, pattern classifier 626 operates with the histograms (FIG. 18) as input. For instance, pattern classifier 626 takes as input the hourly radar echo histogram. Each histogram is constructed from a continuous data stream sampled at approximately 10times per second, where each echo is binned according to its associated distance (x-axis) and reflected energy (y-axis). Over the course of one hour, the system aggregates these readings into a two-dimensional matrix in which each bin contains the total number of echoes exhibiting a specific range of distance and energy. These histograms capture a compressed but information-rich representation of physical activity in the radar field of view, serving as the foundational input for human activity classification decisions.

[0075] Pattern classifier 626 analyzes each histogram for characteristic features that differentiate between human activity and non-human or spurious sources. Human activity is characterized by a broad and dynamic range of distance and energy signatures, reflecting natural movements such as walking, turning, or gesturing, which cause varying radar crosssections and changing distances. In contrast, certain spurious sources produce narrowly concentrated and repetitive echo patterns — such as vibrating window blinds that generate persistent, low-variance reflections at a fixed distance. Similarly, pets tend to generate lower-energy echoes with specific spatial features, including unusually short distances if the pet is on a countertop where the sensing device is positioned. Finally, pattern classifier 626 incorporates additional contextual features to improve accuracy, including time-of-day weighting to reduce the likelihood of false positives during circadian low-activity windows (e.g., 3:00 AM), as well as historical pattern matching to recognize established behavioral norms for the client. This combination of spatial, energetic, temporal, and historical inference enables even more robust discrimination between human and non-human motion signatures in real-world environments than can be provided by the real-time on-device gating mechanisms.

[0076] At daily report time check 628, the server determines whether it is time to generate a daily report. If so, the server performs an activity check 630 on the data to check for alert conditions.

[0077] If the outcome of activity check 630 indicates an alert condition is present, the server initiates transmission 632 of an alert via email, text message, or the like.

[0078] If a daily report is not due during daily report time check 628 or in response to activity check 630, the server determines whether the client has been detected 634 “up and around” for the first time that day. If so, the server transmits sends an “up and around” notification via email, text message, or the like. Otherwise, process 600 proceeds to check the next radar-based monitoring device.

[0079] FIG. 7 shows a process 700 in the form of a flowchart illustrating how the pattern classification can be refined and adapted over time. In this example, process 700 includes ahuman-supervised function 702. Skilled persons will appreciate in light of this disclosure, however, that supervised functions may also be automated (e.g., fully autonomous), in some embodiments.

[0080] Processing begins with triggering periodic review of pattern classifications 704, for example, once daily. The operator reviews 702 a histogram pattern 706 (see, e.g., FIG. 17) for consistency with human activity, pet activity, or other known radar return patterns. Simultaneously the operator also reviews the circadian activity patterns in an advanced sensor analysis screen 708 (see, e.g., FIG. 15).

[0081] The operator then determines whether classification is clear 710 based on histogram data and historical behavioral context. If classification is unclear, the operator initiates contact 712 with the client to confirm whether activity occurred at the relevant time of the data. With this data in hand, the operator may confirm or re-tag the item 714.

[0082] If the item is re-tagged, the operator will then determine 716 if the misclassification can be resolved by a simple adjustment to the real-time gating of distance, energy, and motion. If so, the adjustment is requested 718 and transmitted back to the device as an update for use in the real-time on-device process (FIG. 6).

[0083] If the item is re-tagged and the misclassification cannot be resolved by a simple adjustment of real-time gating settings, a client-specific retraining of the pattern classification system is initiated 720 with the newly corrected tags.

[0084] FIG. 8 illustrates a histogram with three gating mechanisms applied during realtime, on-device decision-making: a minimum distance gating 802, a maximum distance gating 804, and a minimum energy gating 806. These gating thresholds are visualized as overlays on a 2D histogram (see, e.g., histogram pattern 706, FIG. 7) of event data plotted by energy versus distance. Minimum distance gating 802 excludes events that occur too close to the device to be meaningful, which often correspond to sensor noise or insignificant movement. Maximum distance gating 804 excludes events that are too far away to be relevant or reliable for classification. Minimum energy gating 806, shown as a dynamic curve, imposes an energy threshold that decreases with distance, corresponding to the echo energy expected from a human-sized radar reflective cross-section at each distance from the sensor.

[0085] Although not shown in this figure, motion gating operates in a parallel manner using thresholds on measured velocity. These gating mechanisms help the device quickly eliminate events unlikely to correspond to meaningful activity, optimizing processing efficiency and extending battery life in embodiments where a battery is included.

[0086] FIG. 9 illustrates a process 900 for identifying sentinel trends in client behavior based on circadian rhythm features. Process 900 operates on data stored in database 622 and may run independently of the real-time presence classification described in FIG. 6.

[0087] Initially, the system extracts 902 circadian features, including wake-up time, bedtime, and sleep continuity, from accumulated activity reports. These features may be derived from hourly presence classifications and time-stamped activity data.

[0088] The extracted feature values are added 904 to a client-specific database that maintains a rolling history of circadian behavior.

[0089] The system performs a comparison 906 between recent feature values and previously established baseline values to detect deviations indicative of emerging trends.

[0090] The system evaluates whether there is a sentinel trend corresponding to declining 908 sleep continuity. If such a trend is detected, the system proceeds to notify 910 designated well-being partners with an explanation of the concern.

[0091] The system evaluates whether there is a trend of increasing 912 irregularity in the client’s circadian patterns, such as inconsistent sleep / wake timing. If such a pattern is found, the system continues to notify 914 designated well-being partners with an explanation of the concern.

[0092] Process 900 supports higher-level behavioral monitoring by identifying longitudinal changes that may indicate deteriorating health, cognitive issues, or environmental disruption. Notifications generated may complement real-time alerts or inform classifier adaptation strategies.

[0093] FIG. 10 illustrates a flowchart of a server-side classification process 1000 for evaluating radar-based presence data. Process 1000 entails receiving an hourly radar echo histogram (see, e.g., FIG. 18) as input 1002 from radar-based monitoring device 110. Process 1000 transforms the histogram into a three-dimensional point cloud 1004, where each point represents a combination of distance (x), reflected energy (y), and echo count (z).

[0094] Process 1000 then applies a Gaussian blur 1006 to smooth the point cloud into a continuous 3D surface. Process 1000 performs a thresholding operation 1008 by slicing the surface at 10% of the peak echo count to extract an elliptical region referred to as the activity profile (AP).

[0095] Process 1000 computes geometric features 1010 of the AP. In this step, geometric and statistical features are extracted from the processed sensor data to characterize candidate activity patterns. The extracted features include: centroid of the data cluster, providing a spatial reference; tilt of the pattern, indicating orientation or directionality;major axis of the best-fit ellipse, representing the dominant dimension of the activity; and minor axis, capturing the secondary dimension perpendicular to the major axis. These features are used to form a multi-dimensional descriptor of the activity pattern, which is then fed into the classification model to determine whether it represents meaningful activity or not.

[0096] Process 1000 classifies the activity profile 1012 using a k-nearest neighbor (KNN) algorithm that compares the ellipse to a library of pre-tagged AP templates. (See, e.g., FIG. 17-FIG. 19 for examples of template shapes.) Process 1000 then generates a classification result 1014 indicating whether the observed pattern corresponds to human or non-human activity.

[0097] Process 1000 optionally evaluates whether to override a presence determination 1016 previously made by the on-device classification, based on confidence thresholds or contextual considerations. Finally, process 1000 flags ambiguous or misclassified data 1018 for review and potential inclusion in future retraining of the activity profile library. Note that the training phase uses supervised machine learning where an operator double checks the classifications for accuracy, but after that is complete, the classification is automatic.

[0098] FIG. 11 illustrates the override evaluation and adaptive retraining process 1100 that occurs when a discrepancy is identified between an on-device real-time decision (RTD) 1102 and a server-based pattern classification (PC) 1104. Process 1100 begins by comparing the RTD result to the PC result in step 1106. If the two results match, as shown in step 1108, no further action is taken. However, if a mismatch is detected, the PC result overrides the RTD decision, as indicated in step 1110, since the server-based classification is considered more accurate due to its broader contextual and historical inputs.

[0099] When an override occurs, process 1100 hypothesizes that one or more of the RTD’s gating parameters (such as minimum distance, maximum distance, or energy threshold) may have contributed to the incorrect result. In step 1112, it formulates a proposed adjustment to these parameters. The system then re-tests this hypothesis using data from the previous 168 hours, as shown in step 1114, to evaluate whether the adjusted gating values would have yielded more consistent alignment with the PC results. If improvement is observed, indicated in step 1116, the new gating parameters are sent to radar-based monitoring device 110 over the cellular link in step 1118. If no improvement is found, the existing parameters are retained, as indicated in step 1126.

[0100] In addition to this feedback loop, the system performs a weekly review of classification accuracy against expected circadian patterns, as shown in step 1120.Anomalies, such as activity detected at times when the client is typically inactive, are flagged for review. In step 1122, these flagged events are subject to human evaluation, in which an operator may contact the client to verify whether the event reflects actual activity or is a misclassification. Confirmed misclassifications are used to retrain the pattern classifier, refining its activity profile templates for that particular client. In some embodiments, this final step involving human input may also be automated, e.g., by generating push confirmations to the client so as to personalize and improve classification accuracy over time.

[0101] FIG. 12 shows an example of a daily report text message 1200 that is triggered as part of a notification service provided by human presence monitoring system 100. Daily report text message 1200 is transmitted to a designated contact individual to provide routine status updates about a monitored client.

[0102] Daily report text message 1200 includes a type and time of report 1202, which specifies that the message is a daily report and indicates the date and time at which the report was generated. A client status 1204 is also included, indicating the result of the system’s activity analysis — for example, that no alerts were generated and that the client is active for the current day.

[0103] An embedded link 1206 is also included within daily report text message 1200. Embedded link 1206 directs the recipient to a well-being update landing page, which provides additional status information. In some embodiments, the landing page may be accessed without requiring user login or authentication, thereby simplifying access for family members or caregivers. Human presence monitoring system 100 may customize message content and embedded links based on user preferences, alert settings, or historical patterns.

[0104] FIG. 13 illustrates a well-being update page 900 that is accessed through a link delivered in a daily report text message, such as that shown in FIG. 12. This page provides a secure, real-time summary of the monitored client’s status and is intended for viewing by designated contact individuals.

[0105] Displayed prominently is alert status 1302, which conveys whether any alert conditions have been triggered. In this example, system 100 confirms that there are currently no alerts. The page may also include a direct call button, enabling immediate voice contact with the client based on a stored mobile number.

[0106] Time of most recent activity 1304 is provided to show when system 100 last confirmed motion or other presence-based activity. This field may be derived from hourly radar sensing or auxiliary sensor input, depending on system configuration.

[0107] The interface includes buttons 1306 and 1308 for quick access to more detailed views of historical data. Button 1306 links to a daily activity chart view, while button 1308 links to an hourly list view. These allow caregivers to verify patterns of activity and environmental conditions over time.

[0108] Well-being update page 1300 may be accessed without login credentials using a secure tokenized link, enabling convenient and privacy-conscious delivery of client status information to authorized recipients.

[0109] FIG. 14 illustrates a daily chart 1400 that may be accessed through a text message link and presents a visual timeline of the client’s activity across multiple days. Each horizontal row corresponds to a single day and displays activity from midnight through the end of the day in hourly intervals.

[0110] Active hours active hour indicator 1402 are displayed as filled segments, indicating that motion or presence was detected by the monitoring device during those time intervals — for example, from 6 a.m. to 7 a.m. Hourly reports with no activity 1404 are shown using contrasting shading, indicating that system 100 reported data for that hour but did not detect meaningful movement.

[0111] Reports not yet generated 1406 appear at the right end of the daily timeline, typically after 8 p.m., and represent future time intervals for which no radar or environmental data has yet been received.

[0112] Each day also includes a daily temperature range 1408, showing the minimum and maximum indoor temperatures recorded by the monitoring device over the 24-hour period.

[0113] Daily chart 1400 allows authorized users to quickly review historical patterns of client activity and home environment conditions, supporting early detection of behavioral or health-related deviations.

[0114] FIG. 15 illustrates an hourly list 1500 that provides a detailed log of the client’s hourly activity and corresponding environmental data, accessible through a secure text message link.

[0115] Each row of hourly list 1500 corresponds to a one-hour interval, with time segments arranged in reverse chronological order from the most recent hour down to earlier hours. Active hours 1502 are labeled “Active” and indicate that the monitoring device detected human presence during that time window, e.g., from 6 a.m. to 7 a.m.

[0116] Hourly reports with no activity 1504 are shown as rows with no “Active” label, indicating that a report was successfully transmitted but no presence was detected during that hour.

[0117] A midnight marker 1506 is displayed to visually separate calendar days, making it easier for the viewer to interpret multi-day data and transitions from one reporting period to the next.

[0118] Hourly temperature readings 1508 are displayed alongside each interval and represent the ambient indoor temperature measured by the monitoring device’s environmental sensors during that hour.

[0119] Hourly list 1500 provides a high-resolution view of the client’s recent behavior and environment, enabling caregivers or contact individuals to identify anomalies, such as nighttime movement, sustained inactivity, or temperature extremes.

[0120] FIG. 16 illustrates an advanced sensor analysis screen 1600 presented within the system’s web or application interface. This screen is designed for reviewing radar signal data, adjusting on-device classifier parameters, and performing tagging for server-based machine learning refinement.

[0121] Sensor selected 1602 identifies the specific monitoring device currently under review. Client name 1604 is displayed in association with the selected sensor, allowing operators to verify the subject of the data being analyzed.

[0122] On the left side of the interface, a daily chart display area 1606 shows multiple days of activity data. Each row presents a bar graph of activity, similar in format to FIG. 14.Hour selected for analysis 1608 is highlighted within this area and corresponds to the radar histogram displayed on the right side of the screen.

[0123] Histogram of radar echoes 1610 occupies the main portion of the analysis interface. The histogram is structured as a grid of cells organized by distance columns 1612 and energy rows 1614. Each cell in the grid, such as cell 1616, contains the number of radar echoes received during the selected hour that fell within a corresponding energy and distance range.

[0124] Settings for real-time, on-device classifier 1618 are shown below the histogram. These include minimum and maximum gating thresholds, event count thresholds, and communication settings for the device. These parameters influence the behavior of the device’s edge classifier and may be manually adjusted based on observed signal characteristics.

[0125] Tagging for server-based machine learning classifier 1620 is located below the classifier settings. Operators may select whether the signal pattern observed for the selected hour corresponds to actual client activity or inactivity, and optionally submit a manual tag to improve future classifier performance. A control is also provided to retrain the classifier using the tagged data.

[0126] Advanced sensor analysis screen 1600 allows for detailed forensic review of radar data, expert-guided system tuning, and iterative refinement of classifier accuracy based on real-world observations.

[0127] FIG. 17 illustrates a typical radar histogram 1700 for a time interval classified as containing no human activity. Histogram 1700 is organized as a grid defined by energy levels along the vertical axis and distance from the sensor along the horizontal axis.

[0128] A selected hour 1702 identifies the time window as ending at 23:00 on April 18, 2025. This timestamp corresponds to a specific interval within the client’s daily activity log, as shown in earlier figures.

[0129] All echoes are below a minimum energy threshold 1704, meaning that the radar returns for this hour were confined entirely to the lowest energy band (0-9 units). Echoes are also spread across multiple distance bins, indicating weak reflections from distant static surfaces or ambient environmental noise.

[0130] Histogram 1700 serves as a representative pattern for an inactive environment, such as an unoccupied room, and is typically used to confirm a “no activity” classification by the real-time or cloud-based classifier. Similar histograms may be used during system calibration or machine learning retraining to refine presence detection accuracy.

[0131] FIG. 18 illustrates a typical radar histogram 1800 for a time interval classified as containing human activity. Histogram 1800 is structured as a matrix with energy levels on the vertical axis and distance bins on the horizontal axis, consistent with previous histogram formats.

[0132] A selected hour 1802 identifies the interval as ending at 16:00 on April 19, 2025.

[0133] Medium range, high-strength echoes 1804 appear concentrated between 4 and 10 feet from the sensor and across energy levels exceeding 20 units. These radar returns form a characteristic pattern of motion consistent with a human subject moving within the sensing zone.

[0134] Histogram 1800 reflects a signal distribution likely to trigger an “active” classification by the real-time on-device processor or the cloud-based machine learningclassifier. This type of signal pattern may also be used as a training reference during model refinement or classifier validation.

[0135] FIG. 19 illustrates a typical radar histogram 1900 for a time interval in which pet activity was detected. Histogram 1900 is formatted as a two-dimensional matrix of radar returns, with energy levels on the vertical axis and distance bins on the horizontal axis.

[0136] A selected hour 1902 identifies the time window as ending at 05:00 on March 25, 2025.

[0137] Short range echoes from a cat on a countertop 1904 are visible in the upper-left portion of histogram 1900. These echoes appear within close distance bins (e.g., 0-2 feet) and moderate energy levels, consistent with radar reflections from a small animal elevated within the radar’s field of view.

[0138] The pattern shown in histogram 1900 differs from typical human activity patterns (see, e.g., FIG. 18), and may be filtered out by applying minimum distance or energy gating thresholds. This example highlights how human presence monitoring system 100 distinguishes between human and non-human motion to reduce false positives and improve classifier reliability.

[0139] FIG. 20 illustrates a contacts interface screen 2000 within an optional companion application configured to manage a support circle for a monitored individual. Screen 2000 allows an administrator or designated caregiver to view, manage, and communicate with individuals listed as contacts within the monitoring system.

[0140] A scrollable list of contact profiles 2002 is shown, with each profile displaying a contact name, image, communication method, and associated label or role. Profile roles may include terms such as “Circle Center” or “Customer Advocate,” and may also feature custom identifiers such as hashtags.

[0141] Each contact profile 2002 includes a set of communication buttons 2004, which may initiate phone calls, text messages, or emails directly from the interface. An edit contact button 2006 is provided to update the individual's contact information or role designation.

[0142] At the top of screen 2000, a search bar 2008 is available to filter the contact list by name or keyword. This feature facilitates navigation in cases where numerous contact entries are stored.

[0143] Toward the bottom of screen 2000, interface controls are presented for administrative tasks. An add contact button 2010 allows a new contact profile to be added tothe support circle. A manage circle button 2012 provides access to higher-level management functions such as reordering, removing, or modifying contact permissions.

[0144] Screen 2000 may be accessed through a mobile or web-based version of the companion application, and is linked to records stored in the system’s secure cloud database.

[0145] FIG. 21 illustrates a well-being status interface screen 2100 in an optional companion application, configured to display current status information for a monitored contact. Screen 2100 is intended for viewing by members of the contact’s support circle, such as family members, caregivers, or system administrators.

[0146] A contact profile section 2102 includes a photo, name, and current sensor location associated with the monitored individual. A change location button 2104 is provided to update the reported position of the monitoring device, particularly when multiple sensors are installed within a residence.

[0147] A call button 2106 allows the user to initiate a voice connection with the contact directly from the interface. Below this, an alert summary 2108 provides real-time feedback on whether any alert conditions are currently active.

[0148] Presence sensor status 2110 shows whether radar-based sensing is active for the monitored individual. A toggle button 2112 allows authorized users to disable or re-enable sensing remotely. An away dates field 2114 displays any date ranges during which the client is expected to be absent from home, and a calendar button 2116 enables users to configure this setting.

[0149] At the bottom of screen 2100, last recorded activity time and maximum temperature for the current day are shown in summary fields 2118 and 2120, respectively. Interface buttons 2122 and 2124 provide quick access to a daily chart view and hourly list view, as previously described with reference to FIG. 14 and FIG. 15.

[0150] Screen 2100 may be viewed via mobile or desktop platforms, and is synchronized with cloud-based system records to ensure up-to-date status and control.

[0151] FIG. 22 illustrates a daily report message setup screen 2200 within an optional companion application, enabling users to configure automated well-being notifications for one or more contacts.

[0152] A time selection interface 2202 allows the user to specify the daily delivery time for well-being reports. A recipient management panel 2204 enables selection of one or more individuals who will receive the reports. New recipients can be added using a dropdown menu and associated controls.

[0153] Each configured recipient is displayed within a recipient card 2206. Recipient card 2206 includes options to enable or disable communication via email, text message, or both. Users may also opt to include a direct link to view a daily activity chart, similar to the example shown in FIG. 14.

[0154] An optional checkbox within recipient card 2206 allows the user to suppress message delivery when no alert conditions are present. This helps reduce notification fatigue while still ensuring that important events are communicated.

[0155] At the bottom of screen 2200, interface controls include a discard changes button 2208, a save changes button 2210, and a send test messages button 2212. These controls provide administrative flexibility and allow users to validate the setup before enabling daily notifications.

[0156] Screen 2200 supports configuration of contact-specific delivery methods and conditions, enabling tailored communication that aligns with each recipient’s preferences and technical capabilities.

[0157] FIG. 23 illustrates an "Up and Around" message setup screen 2300 within an optional companion application. Screen 2300 allows authorized users to configure automatic notifications that are sent when the monitored individual is first detected as active for the day.

[0158] A delivery condition panel 2302 defines the rules for message transmission, including a selected time threshold before which no message will be sent. This ensures that notifications are not triggered by early-morning movement outside of the user’s expected schedule.

[0159] Recipient management tools 2304 allow the user to select and manage individuals who will receive "Up and Around" notifications. Each recipient is listed in a recipient card 2306, which includes contact methods, such as email and SMS text message, along with delivery options.

[0160] Each recipient card 2306 includes options to enable or disable delivery to a specific email address or phone number, as well as a setting to include a direct link to view a daily activity chart, consistent with the format shown in FIG. 14.

[0161] At the bottom of screen 2300, user interface controls include a discard changes button 2308, a save changes button 2310, and a combined save and send test messages button 2312. These controls allow the user to validate and confirm the setup before activating the notification logic.

[0162] Screen 2300 facilitates proactive communication with a client’s care circle, allowing family members or caregivers to be informed promptly when the client is up and moving, thereby reinforcing daily engagement and providing peace of mind.

[0163] FIG. 24 shows a process 2400 for monitoring human presence in a residential environment. In block 2402, process 2400 detects radar reflections at a presence-sensing device. In block 2404, process 2400 classifies motion locally as human or non-human based on distance and energy gating thresholds. In some embodiments, classifying motion includes applying gating logic that incorporates minimum and maximum thresholds for distance, motion, and signal energy.

[0164] In block 2406, process 2400 aggregates radar signals into a compressed histogram format. In some implementations, this includes generating a two-dimensional histogram of radar returns organized by distance and energy bins. Radar reflections may be sampled at a rate of approximately ten times per second. The histogram may be compressed into a report of approximately 256 bytes for efficient transmission over a low-bandwidth link because the histogram is 10x 10, i.e., 100 distance / energy bins; each bin holds a two-byte number (0- 65535) which is the total hits in that bin for that hour. Combined that is 200 bytes for transmission.

[0165] In block 2408, process 2400 transmits the compressed histogram to a cloud-based server over a cellular connection. In block 2410, the server analyzes the histogram data using a classifier trained on historical activity and user-specific circadian rhythm models. In some embodiments, the classifier is trained on both historical presence data and individualized circadian rhythm patterns derived from prior behavior. The classifier may also be updated based on operator-tagged activity data received from a review interface.

[0166] In block 2412, the server determines if local classification thresholds require adjustment. If so, in block 2414, the server transmits updated thresholds to the presencesensing device without requiring manual configuration or user input.

[0167] In some embodiments, updated thresholds are used to automatically adjust on- device gating logic. Other optional enhancements to process 2400 may entail activating a visual indicator that shows a first indication (e.g., green) when human presence is detected and server connectivity is intact, and a second indication (e.g., red) if connectivity is lost; declaring an activity event locally when motion persistence exceeds a configured threshold; transmitting a periodic report that includes hourly presence status, histogram data, and temperature readings; or transmitting a notification to a designated contact individual if the server determines that an alert condition is met.

[0168] FIG. 25 is a block diagram illustrating hardware components 2500 that may be used to implement monitoring device 110, server functions 104, or both, according to various embodiments. Example computing implementation 2502 include one or more processors 2504, one or more memory / storage devices 2506, and one or more communication resources 2508. These resources may be communicatively coupled via one or more data buses 2510.

[0169] Processors 2504 may include, for example, a central processing unit (CPU), graphics processing unit (GPU), digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other specialized processing unit, or any combination thereof. As illustrated, processors 2504 may include processor 2512 and processor 2514, each capable of executing instructions 2516.

[0170] Memory / storage devices 2506 may include volatile and non-volatile memory components, such as dynamic random-access memory (DRAM), static RAM (SRAM), Flash memory, solid-state storage, hard disk drives, or any combination thereof. Instructions 2516 may reside in memory / storage devices 2506 for execution by one or more processors 2504.

[0171] Communication resources 2508 enable example computing implementation 2502 to transmit data to and receive data from external components. These may include wired and wireless communication interfaces such as USB, Ethernet, Wi-Fi, Bluetooth®, Zigbee, cellular modems, or other network interfaces. Communication resources 2508 may facilitate interaction with peripheral devices 2518, a network 2520, or one or more databases 2522, all of which may also store or access instructions 2516.

[0172] Instructions 2516 may include software, firmware, or executable code configured to perform one or more of the methods described herein. These methods may include radar signal processing, presence classification, histogram compression, activity reporting, machine learning-based model training, alert generation, and related functions as described with reference to FIG. 1-FIG. 24.

[0173] Instructions 2516 may be stored in whole or in part in memory / storage devices 2506, within processors 2504 (e.g., in a processor cache), in peripheral devices 2518, or in remote databases 2522. Accordingly, any of these components may constitute or access a non-transitory machine-readable medium storing instructions for performing the disclosed techniques.

[0174] In light of this disclosure, skilled persons will appreciate that many changes may be made to the details of the above-described embodiments without departing from the underlying principles of the invention. The scope of the present invention should, therefore, be determined only by claims and equivalents.

Claims

CLAIMSWhat is claimed is:

1. A human presence monitoring system for use in a residential environment, comprising: a radar-based monitoring device, including: a radar sensor configured to emit frequency-modulated continuous-wave radar signals and receive reflected signals; processing circuitry configured to classify motion as indicative or nonindicative of human presence using distance, energy, and motion gating; generate a presence indication signal upon classifying motion as indicative of human presence; activate a visual indicator that provides a first indication of both human presence and end-to-end system connectivity, and provides a second indication when connectivity to the server is lost; accumulate radar data into a two-dimensional histogram based on distance and signal energy; compress the histogram into a compact data format; and a cellular communication module configured to periodically transmit the compressed data; a remote server configured to analyze received compressed data using a pattern classifier trained on historical activity and user-specific circadian rhythms; determine whether to override the on-device classification; and transmit updated classification parameters to the device automatically, without human intervention.

2. The human presence monitoring system of claim 1, in which the radar-based monitoring device further comprises a housing configured to be mounted on a power outlet.

3. The human presence monitoring system of claim 1, in which the radar-based monitoring device further comprises a housing configured to be supported by a stand and powered via a cable and power supply.

4. The human presence monitoring system of claim 1, in which the radar data is accumulated into a histogram comprising distance bins along a first axis and signal energy bins along a second axis.

5. The human presence monitoring system of claim 1, in which the radar-based monitoring device samples radar signals at a rate of approximately ten times per second.

6. The human presence monitoring system of claim 1, in which the visual indicator is configured to illuminate green in response to detecting human presence with server connectivity, and red in response to loss of connectivity.

7. The human presence monitoring system of claim 1, in which the processing circuitry is configured to apply gating rules comprising a combination of minimum and maximum values for distance, motion, and energy.

8. The human presence monitoring system of claim 1, in which the processing circuitry is configured to transmit activity reports to the server at fixed hourly intervals.

9. The human presence monitoring system of claim 1, in which the histogram data is compressed into a format of no more than 256 bytes per hourly report for transmission over a low-bandwidth cellular channel.

10. The human presence monitoring system of claim 1, further comprising a rechargeable battery configured to support continued operation of the device in the event of power loss.

11. The human presence monitoring system of claim 1, in which the server applies a model of client-specific circadian rhythms to determine whether a received activity report deviates from expected behavior.

12. The human presence monitoring system of claim 1, in which alert conditions are adjusted over time based on the client’s observed daily routine.

13. The human presence monitoring system of claim 1, in which the radar-based monitoring device is configured to operate without requiring any direct physical interaction from the monitored individual.

14. The human presence monitoring system of claim 1, in which the housing of the radarbased monitoring device is styled to resemble a night light or picture frame for integration into a residential environment.

15. The human presence monitoring system of claim 1, in which the server is further configured to update a machine learning model based on operator-labeled presence data.

16. The human presence monitoring system of claim 1, in which the server adjusts gating thresholds for the radar-based monitoring device based on user-specific activity feedback.

17. The human presence monitoring system of claim 1, in which updated classification parameters transmitted to the device are based on data manually tagged by a human reviewer.

18. The human presence monitoring system of claim 1, in which the server is configured to transmit a daily status report to a designated contact individual at a preselected time.

19. The human presence monitoring system of claim 1, in which the server is further configured to transmit a message upon detecting first activity for the day.

20. The human presence monitoring system of claim 1, further comprising a companion application operable on a mobile device or web browser, configured to display activity history and configure contact preferences.

21. The human presence monitoring system of claim 20, in which the companion application displays a well-being status page including most recent activity, alert status, and ambient temperature.

22. A presence-sensing device, comprising: a radar sensor configured to emit and receive millimeter-wave radar signals; processing circuitry configured to apply motion, distance, and energy-based gating to identify likely human presence; aggregate radar signal returns into a distance-energy histogram; compress the histogram and transmit it over a wireless modem; and receive updated classification parameters from a remote server trained on historical behavior patterns.

23. The device of claim 22, in which the processing circuitry samples radar signal data at a rate of approximately ten times per second.

24. The device of claim 22, in which the distance-energy histogram comprises a two- dimensional matrix of radar returns binned by distance along a first axis and signal energy along a second axis.

25. The device of claim 22, in which the processing circuitry is further configured to declare an activity event when the histogram meets configurable gating thresholds for energy, distance, and motion persistence.

26. The device of claim 22, in which the processing circuitry is further configured to determine an hourly presence status based on a count of activity events within a reporting interval.

27. The device of claim 22, in which the processing circuitry is further configured to cause a status indicator to display a first indication when human presence is detected and the device is connected to a server, and a second indication if connectivity is lost.

28. The device of claim 27, in which the status indicator comprises an RGB LED configured to display different colors corresponding to device status.

29. The device of claim 22, in which the wireless modem transmits a compressed data payload of less than 256 bytes per hourly report.

30. The device of claim 22, further comprising a rechargeable battery configured to maintain device operation during a power failure.

31. The device of claim 22, further comprising a temperature and humidity sensor, and in which the processing circuitry is configured to include environmental sensor readings in the compressed data.

32. The device of claim 22, in which the device housing is configured to be mounted directly into a wall outlet or supported by a countertop stand.

33. The device of claim 22, in which the device is configured to operate without requiring any physical interaction from the user.

34. A method for monitoring human presence in a residential environment, comprising: detecting radar reflections at a presence-sensing device; classifying motion locally as human or non-human based on distance and energy gating thresholds; aggregating radar signals into a compressed histogram format; transmitting the histogram to a cloud-based server over a low-bandwidth cellular link; analyzing the histogram data at the server using a classifier trained on historical activity and user-specific circadian rhythm models; determining if local classification thresholds require adjustment; andtransmitting updated thresholds to the presence-sensing device without requiring manual configuration.

35. The method of claim 34, further comprising activating a visual indicator that shows a first indication when human presence is detected and server connectivity is intact, and a second indication if connectivity is lost.

36. The method of claim 34, in which classifying motion comprises applying gating logic that includes minimum and maximum thresholds for distance, motion, and signal energy.

37. The method of claim 34, in which aggregating radar signals comprises generating a two- dimensional histogram of signal returns organized by distance and energy bins.

38. The method of claim 34, in which the radar reflections are sampled at a rate of approximately ten times per second.

39. The method of claim 34, in which transmitting the histogram comprises compressing the data into a report not exceeding 256 bytes.

40. The method of claim 34, further comprising declaring an activity event on the presencesensing device based on motion persistence exceeding a configured threshold.

41. The method of claim 34, further comprising transmitting a periodic report from the presence-sensing device that includes at least the hourly activity status, compressed histogram data, and temperature readings.

42. The method of claim 34, in which the classifier is trained on both historical presence data and individualized circadian rhythm patterns derived from prior behavior.

43. The method of claim 34, further comprising receiving operator-tagged activity data at the server to retrain the classifier.

44. The method of claim 34, further comprising transmitting a notification to a contact individual if the classifier determines that an alert condition has been met.

45. The method of claim 34, in which transmitting updated thresholds comprises automatically adjusting on-device gating logic without requiring user input or physical access.

46. The method of claim 34, further comprising extracting circadian rhythm features from radar-derived presence data, the features including at least one of: wake-up time, bedtime, or sleep continuity.

47. The method of claim 34, further comprising comparing recent circadian features to baseline values to detect deviations indicative of sentinel trends.

48. The method of claim 34, further comprising identifying a sentinel trend corresponding to declining sleep continuity or increasing irregularity of circadian patterns.

49. The method of claim 34, further comprising transmitting a notification to a designated contact individual if a sentinel trend is detected, the notification including an explanation of the detected deviation.

Images