A millimeter wave radar-based life perception pattern recognition method and system

Abstract

The application provides a millimeter wave radar-based life perception mode recognition method and system, relates to the technical field of life perception, and has the technical solution points that monitoring point information of multiple body parts of a user object and vital sign monitoring information are acquired; in the case where it is determined that the body of the user object has a slow and continuous downward trend, the motion trajectory of the body of the user object is analyzed, in the case where a relative parameter meets a preset abnormal relative parameter condition, an abnormal stability change mark is triggered, and in the case where it is determined that the body position of the user is abnormal and an abnormal stationary state of the body of the user object is determined, an abnormal low-position stationary mark is triggered; a dangerous event warning is triggered based on the abnormal stability change mark and the abnormal low-position stationary mark. The application aims to realize accurate perception of a slow motion dangerous event, and can identify an abnormality through multi-dimensional logical judgment in an environment where a signal is weak or deviates, so that misjudgment caused by feature loss is avoided.

Description

Technical Field

[0001] This application relates to the field of life sensing technology, and more specifically, to a life sensing pattern recognition method and system based on millimeter-wave radar. Background Technology

[0002] Millimeter-wave radar life sensing systems can assess the condition of the elderly by emitting millimeter waves and analyzing the minute motion characteristics of the reflected signals from the human body, enabling non-contact monitoring of the elderly's vital signs and behavioral patterns. However, during long-term operation, physical disturbances such as environmental cleaning and equipment inspections can cause minor structural changes in the antenna array, such as slight angle drops or shifts. Although these changes do not trigger hardware failures, they can cause the radar's main beam to deviate from the core monitoring area, leading to attenuation of the reflected signals from the target human torso. Meanwhile, the relative intensity of background clutter generated by low furniture increases, resulting in a deterioration in the signal-to-interference ratio. This deterioration in signal quality makes it difficult for the system to reliably extract weak vital sign data. Furthermore, when the elderly experience atypical motion-related dangerous events, insufficient feature extraction and signal masking can lead to high-risk events being incorrectly identified as posture adjustments or static states, causing the early warning function to completely fail. This can result in missed detections and endanger the lives of the elderly. Summary of the Invention

[0003] The purpose of this application is to provide a life perception pattern recognition method and system based on millimeter-wave radar, which aims to achieve accurate perception of slow-moving dangerous events and identify anomalies through multi-dimensional logical judgment in environments with weak signals or deviations, thereby avoiding misjudgments caused by missing features.

[0004] Firstly, this application provides a life-sensing pattern recognition method based on millimeter-wave radar, including: Obtain body monitoring information of a user's body based on millimeter-wave radar monitoring. The body monitoring information includes monitoring point information of multiple body parts of the user and vital sign monitoring information. When it is determined through monitoring information from multiple body parts that the user's body has a slow and continuous downward trend, the movement trajectory of the user's body is analyzed through the body monitoring information. If the difference between the movement trajectory and the preset normal posture change trajectory is within the preset trajectory range, a non-voluntary posture change marker is triggered. Based on the non-autonomous posture change marker, relative parameters between different body parts are obtained through monitoring point information of different body parts. When the relative parameters meet the preset abnormal relative parameter conditions, an abnormal stability change marker is triggered. When the abnormal body position of the user is determined through the monitoring point information of the multiple body parts, and the abnormal static state of the user's body is determined through the vital sign monitoring information and the monitoring point information of the multiple body parts, an abnormal low-position static marker is triggered. Based on the aforementioned abnormal stability change markers and abnormal low-level stationary markers, a danger event warning is triggered.

[0005] Secondly, this application provides a life-sensing pattern recognition system based on millimeter-wave radar, comprising: The information acquisition module is used to acquire body monitoring information of the user's body obtained based on millimeter-wave radar monitoring. The body monitoring information includes monitoring point information of multiple body parts of the user and vital sign monitoring information. The abnormal trajectory module is used to analyze the user's body movement trajectory through the body monitoring information when it is determined that the user's body has a slow and continuous downward trend through the monitoring information of multiple body parts. If the difference between the movement trajectory and the preset normal posture change trajectory is within the preset trajectory range, a non-autonomous posture change marker is triggered. The abnormal state module is used to obtain relative parameters between different body parts based on the non-autonomous posture change marker through monitoring point information of different body parts. When the relative parameters meet the preset abnormal relative parameter conditions, it triggers the abnormal stability change marker. When it is determined that the user's body position is abnormal through the monitoring point information of multiple body parts, and the abnormal static state of the user's body is determined through vital sign monitoring information and monitoring point information of multiple body parts, it triggers the abnormal low-position static marker. The early warning module is used to trigger a warning of dangerous events based on the abnormal stability change marker and the abnormal low-level stationary marker.

[0006] As can be seen from the above, the life perception pattern recognition method based on millimeter-wave radar provided in this application achieves in-depth analysis of human behavior and posture by integrating body part monitoring information, vital sign monitoring information, and auxiliary environmental feedback information. Furthermore, addressing the signal quality degradation and weak feature issues caused by radar antenna offset, it improves the accuracy of lower body monitoring data through predictive search area and interpolation optimization strategies. Simultaneously, by introducing multi-level logical judgments using non-autonomous posture change markers, abnormal stability markers, and abnormal low-position stationary markers, it can accurately identify difficult-to-identify atypical motion hazard events such as slow slips. This allows for the linked analysis of physical environmental constraints and human behavioral characteristics, solving the problem of early warning failure caused by signal interference and ensuring the safety of the elderly in elderly care environments. Attached Figure Description

[0007] Figure 1 This is a flowchart of a life perception pattern recognition method based on millimeter-wave radar provided in one embodiment of this application; Figure 2 This is a flowchart of a life perception pattern recognition method based on millimeter-wave radar provided in another embodiment of this application; Figure 3 This is a flowchart of a life perception pattern recognition method based on millimeter-wave radar provided in another embodiment of this application. Detailed Implementation

[0008] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0009] It should be noted that similar reference numerals and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0010] Firstly, such as Figure 1 As shown, Figure 1 This is a flowchart of a life perception pattern recognition method based on millimeter-wave radar provided in one embodiment of this application. The method may include, but is not limited to, steps S110 to S140.

[0011] Step S110: Obtain body monitoring information of the user's body based on millimeter-wave radar monitoring. The body monitoring information includes monitoring point information of multiple body parts of the user and vital sign monitoring information. Step S120: When it is determined through monitoring information of multiple body parts that the user's body has a slow and continuous downward trend, the motion trajectory of the user's body is analyzed through body monitoring information. If the difference between the motion trajectory and the preset normal posture change trajectory is within the preset trajectory range, a non-voluntary posture change marker is triggered. Step S130: Based on the non-autonomous posture change marker, the relative parameters between different body parts are obtained through the monitoring point information of different body parts. When the relative parameters meet the preset abnormal relative parameter conditions, the abnormal stability change marker is triggered. When the abnormal body position of the user is determined through the monitoring point information of multiple body parts, and the abnormal static state of the user's body is determined through the vital sign monitoring information and the monitoring point information of multiple body parts, the abnormal low-position static marker is triggered. Step S140: Based on the abnormal stability change marker and the abnormal low-level static marker, trigger the danger event warning.

[0012] For example, millimeter-wave radar can acquire multi-dimensional body monitoring information, including monitoring point information for multiple body parts of the user and vital sign monitoring information. Body monitoring information refers to multi-dimensional state data about the target object extracted from reflected signals by the millimeter-wave radar system, including physical coordinate information representing the spatial position of various body parts and micro-motion information representing internal physiological states. The monitoring point information for multiple body parts is obtained from continuously acquired radar point clouds. After target separation, clustering, and temporal correlation, it corresponds to one or more sets of point information for the head, neck and shoulder region, trunk region, upper limb region, hip region, and lower limb region, respectively. Each set of point information can at least represent the spatial position of the body part at the current moment, the distribution range of the point cloud, and the positional change relationship with previous and subsequent moments. Vital sign monitoring information includes micro-displacement changes in the chest, abdomen, or trunk region, which can be represented as a continuously changing micro-motion sequence over time. After filtering, denoising, and periodic analysis, respiratory-related features, cardiac-related features, or state-related results such as weak vital signs or unstable vital signs can be obtained.

[0013] When monitoring data from multiple body parts indicates a slow and continuous downward trend in a user's body, it is not immediately classified as a fall. Instead, the movement trajectory of the user's body, analyzed through body monitoring information, is compared with a normal posture trajectory to identify an involuntary slip. If the difference between the movement trajectory and the preset normal posture change trajectory is within a predetermined range, an involuntary posture change marker is triggered. This marker characterizes a gradual change in body posture that is not controlled by conscious will, such as a slow, limp slip due to loss of muscle control. The involuntary posture change marker does not require the target to experience a sudden, drastic acceleration; rather, it emphasizes the continuity, directionality, and loss of control of the posture change. For example, a continuous decrease in overall body height, weakened trunk support, and a posture trajectory deviating from the pattern of normal controlled movement.

[0014] Subsequently, relative parameters between different body parts are obtained through monitoring points on different body parts. Further analysis of these relative relationships confirms whether the body has lost its mechanical stability. Vital signs monitoring information and monitoring point information from multiple body parts, combined with spatial location and vital signs, confirms whether the user is in a dangerous static state, triggering an abnormal low-position static marker. This marker indicates that the target object, after undergoing a posture change, is in an abnormally close-to-the-ground spatial position and lacks normal physiological and physical activity characteristics, indicating a dangerous state. Here, "low-position" does not refer to an absolute low height, but rather that the target object has entered a near-ground area where it should not remain for an extended period, relative to common activity areas such as normal standing, sitting, lying down, or wheelchair use. "Static" here does not mean completely devoid of point cloud changes, but rather that during the continuous observation period, apart from weak noise disturbances, the main body parts do not exhibit displacement changes consistent with conscious activity, and vital signs are abnormal or significantly weakened.

[0015] Therefore, the concept of this application's embodiments is a progressively tightening judgment chain. The first layer focuses on whether a continuous descent is occurring, used to filter out suspicious processes from a large number of daily activities. The second layer focuses on whether the descent process is closer to an uncontrolled slide than a controlled action, used to avoid direct false alarms for normal actions such as sitting, squatting, and bending over. The third layer focuses on whether the internal geometric relationship of the body is unstable, used to confirm that the process is not a simple posture change, but is accompanied by the collapse of the supporting structure or abnormal shift of the center of gravity. The fourth layer focuses on whether it eventually enters an abnormally low position and remains stationary, used to distinguish between a brief imbalance and a dangerous state. A dangerous event warning is only output when multiple layers corroborate each other. Therefore, this application's embodiments do not rely on a single feature, but on the fusion of multiple evidences in a continuous process, thereby adapting to scenarios with slight radar angle deviation, enhanced clutter, and weakened echo.

[0016] In one embodiment, a millimeter-wave radar with a configured multi-input multi-output antenna array can transmit frequency-modulated continuous waves. After receiving the reflected signal, the millimeter-wave radar performs fast Fourier transform processing to generate range-angle-Doppler three-dimensional point cloud data. By performing density clustering algorithms on the point cloud data, monitoring point information representing multiple body parts such as the head, torso, and limbs can be separated. At the same time, for the monitoring points in the torso region, phase dewinding technology is used to extract minute periodic phase changes, thereby obtaining vital sign monitoring information such as breathing and heartbeat. In this way, the spatial contour of the human body can be initially constructed by using basic point cloud clustering.

[0017] For example, millimeter-wave radar can be installed on the wall or ceiling corner of a room, covering the area near the bed, activity area, and floor. After the millimeter-wave radar continuously transmits frequency-modulated continuous waves, the receiver obtains echo signals from multiple receiving channels. First, it distinguishes different range units in the distance dimension, and then distinguishes reflectors in different directions in the angle dimension. Combined with Doppler information, it identifies the motion state of the target.

[0018] For single-person monitoring in elderly care scenarios, static background modeling can be used to remove long-term stable reflections from walls, cabinets, beds, etc., and then connectivity analysis and density clustering can be performed on the remaining point cloud to obtain the main point cloud cluster of the human body.

[0019] For the main point cloud clusters of the human body, secondary segmentation can be performed based on height distribution, point cloud density center, and motion consistency across consecutive frames. Point clusters located at higher positions with relatively concentrated point counts can be considered as candidate head regions; regions located in the middle with relatively stable point clouds and strongest correlation with respiratory micro-movements can be considered as candidate torso regions; and point clusters located in the lower part that significantly migrate with posture changes can be considered as candidate lower limb regions. If point clouds of certain parts are temporarily missing at a certain moment, compensation and correlation can be performed by combining the trajectory continuity of previous and subsequent moments, that is, using the part labels already confirmed at the previous moment and updating them based on the neighboring positions of the current point cloud.

[0020] For acquiring vital sign monitoring information, several monitoring points with stable echoes can be selected within the candidate area of ​​the torso to track subtle changes in phase over time. First, low-frequency drift caused by large-amplitude body movements is filtered out, and then respiratory fluctuation features are extracted from the remaining micro-movements. When the signal quality allows, cardiac-related features are further extracted from more subtle periodic changes. If the cardiac features are unstable, at least the presence of respiration and the trend of changes in respiratory strength are retained as part of the vital sign monitoring information.

[0021] For example, when judging a slow and continuous downward trend, the height center of the overall monitoring point cloud can be calculated in real time. When the height center continuously decreases within a set time window and the rate of decrease is lower than the abrupt change rate threshold of a typical fall, a slow downward trend is confirmed. At this time, the motion trajectory of this process can be extracted, and a dynamic time warping algorithm can be used to compare its similarity with a preset normal posture change trajectory. If the calculated difference value is within the preset trajectory range, it indicates that the motion trajectory is different from normal motion in terms of macroscopic trajectory. For example, if the similarity between the current trajectory and the preset normal trajectory features is lower than a certain threshold or the deviation is higher than a certain threshold, a non-voluntary posture change marker is triggered, thereby identifying a slow slip event caused by weakness.

[0022] The height center of the overall monitoring point cloud can be obtained by comprehensively measuring the height of monitoring points on multiple body parts such as the head, torso, hips, and lower limbs. The torso and hips have higher reference weights because they better represent the main body position and are less affected by local arm swings, hand raisings, and other movements during the slow descent. Continuous descent does not require strict descent in every frame, but allows for slight rebounds or noise fluctuations at local moments. As long as the overall trend is downward within the continuous observation interval and the descent process has temporal continuity, it can be considered a slow and continuous descent trend.

[0023] Furthermore, to avoid misidentifying normal bending over as a downward trend, when extracting the movement trajectory, the overall height change trajectory and the posture change trajectory of the trunk area relative to the lower limb area can be recorded simultaneously. Normal bending over is characterized by a significant forward tilt of the upper body, but limited changes in hip height. Normal sitting down is characterized by a gradual descent of the hips and eventually stopping at a stable sitting height. Involuntary slipping is characterized by a continuous downward shift in overall height, weakened trunk support, unstable stopping point, and a tendency to enter an abnormally low position after the descent ends.

[0024] Preset normal posture change trajectories can be obtained by pre-collecting samples of common daily actions in elderly care scenarios. For example, continuous trajectory data of actions such as sitting down normally, slowly getting up from the bedside and sitting back down after failing, bending over to pick up an object, and squatting down to organize items can be collected, classified, and organized to form a corresponding trajectory template library. When establishing this trajectory template library, each type of action is collected multiple times, and then the trajectories of similar actions are time-aligned and morphologically summarized to retain common change features, and finally, preset normal posture change trajectories that can be used for comparison are obtained.

[0025] For example, after triggering a non-voluntary posture change marker, the basic spatial straight-line distance between monitoring points of different body parts is calculated as a relative parameter. If this distance exceeds the range allowed by normal physiological structure, an abnormal stability change marker is triggered. At the same time, it is determined whether the overall point cloud height is lower than the set safety plane to determine the positional anomaly, and combined with the degree of weakness of vital signs, an abnormal static state is preliminarily judged. The basic spatial straight-line distance between different body parts can include at least the distance between the head and trunk, the distance between the trunk and hip, the distance between the hip and the main point cluster of the lower limb, and the distance between the head and the main point cluster of the lower limb. It is not necessary to determine whether the distance is completely consistent with the actual anatomical length of the human body, but rather to focus on whether the changes in the continuous time sequence suddenly lose the normal proportional relationship. For example, during normal sitting, although the relative distances between the head, trunk, hip, and lower limb may change, the overall human structure topology still maintains a relatively stable structure. However, during flaccid slippage, the head may quickly approach the ground, the relative positional relationship between the trunk and hip may be abnormally compressed or shifted, and the lower limb may remain in place or move with lag, thus causing the distance relationship between multiple parts to be abnormal at the same time.

[0026] A safe plane can be a reference plane used to distinguish between normal activity height and near-ground danger height in the current scene. After the equipment is installed, point clouds of the open environment can be collected first, and the ground primary reflector layer can be identified as the ground reference. Then, combined with the height distribution of support surfaces such as bed surface, seat surface, and wheelchair seat, a low-level reference interval for hazard identification can be defined to obtain the safe plane. When the human body point cloud enters the low-level reference interval for a long time and does not correspond to the known normal support surface dwell pattern, it can be determined as an abnormal position.

[0027] For initial assessment of abnormal static states, one can first observe the stability of vital sign monitoring information across several consecutive frames. For example, whether respiratory micro-movements continuously weaken, whether effective periodic features are difficult to extract for extended periods, and whether major body parts lack active adjustment movements. By using an initial screening followed by continuous confirmation, misjudgments caused by missing point clouds in a single frame, local occlusion, or enhanced clutter can be reduced.

[0028] Understandably, when the antenna angle is slightly off, the echo intensity and local point cloud integrity are affected first, rather than the overall movement trend of the human body over a longer time scale. Therefore, if the identification mechanism mainly relies on instantaneous strong echoes or violent velocity peaks, it is prone to failure. However, the embodiments of this application first capture the cumulative time features of continuous descent. Even if the quality of a single frame fluctuates, as long as sufficient main displacement information is retained in consecutive frames, the descent trend can be recovered. Secondly, trajectory comparison does not require the trajectory to be completely consistent with the dangerous template, but rather to compare it with the trajectory of normal posture change, thereby identifying processes that appear normal but are actually out of control. This reverse identification approach is more suitable for slow descent scenarios. In addition, the abnormal stability change marker focuses on the relative relationship between different parts of the body. The relative relationship is not sensitive to changes in the overall echo amplitude. Even if the overall signal weakens, as long as the relative positions of the main parts can still be extracted, it can still be determined whether structural instability has occurred. The duration dimension of the abnormal low-position stillness marker will not immediately trigger an alarm due to brief low-position movements or failure to extract instantaneous vital signs. Instead, it requires that low position, stillness, and abnormal vital signs are all established within a certain duration.

[0029] In other words, this application does not use a rapid fall detection mechanism that relies on strong Doppler frequency shift. Instead, it utilizes a slow and continuous downward trend combined with trajectory comparison to identify atypical slippage processes. Thus, even in environments with enhanced background clutter and attenuated effective signals, it can still extract the characteristics of involuntary slippage from blurred signals through long-term trajectory morphology analysis. Furthermore, by comprehensively considering the relative parameters of body parts, positional anomalies, and abnormal static states of vital signs through multiple cross-validation, the error tolerance rate in harsh signal environments is improved, ensuring stable extraction and accurate early warning of slow, dangerous events in the elderly.

[0030] The flowchart of a life perception pattern recognition method based on millimeter-wave radar provided in one embodiment of this application includes relative parameters such as relative angle and relative distance; regarding the above step S130, the method may include, but is not limited to, steps S210 to S220.

[0031] Step S210: Obtain the relative angles and distances between different body parts, as well as the center of gravity monitoring information, through the monitoring point information of different body parts; Step S220: If the rate of change of the relative angle meets the corresponding preset abnormal relative parameter conditions, trigger the abnormal stability change flag; or, if the relative distance between the center of gravity monitoring information and the monitoring point information of different body parts, as well as the relative distance between the monitoring point information of different body parts, both meet the corresponding preset abnormal relative parameter conditions, trigger the abnormal stability change flag.

[0032] For example, by weighting the monitoring point information of each body part, the monitoring information of the human body's center of gravity can be estimated in real time. At the same time, the relative angle between the trunk and the lower limbs, as well as the relative distance between the head and the center of gravity, can be calculated. If an elderly person experiences flaccid paralysis, the relative angle between the trunk and the lower limbs will change drastically in a short period of time, or the center of gravity will rapidly deviate from the center of the trunk, resulting in an abnormal distribution of the relative distance between the center of gravity and each part. As long as any of the above geometric relationship mutation conditions are met, an abnormal stability change marker can be triggered.

[0033] The relative angle can be obtained by selecting the main direction of two adjacent body parts. For example, for the torso region, the main direction of the torso can be determined by the extension direction of the torso point cluster in consecutive frames. For the lower limb region, the main direction of the lower limb can be determined by the direction of the line connecting the hip to the main point cluster of the lower limb. The change in the angle between the torso region and the lower limb region can be used as the relative angle between the torso and the lower limb.

[0034] Relative distance can be represented by the spatial interval between the center point, principal point, or stable reflection point of each body part. The center of gravity monitoring information is an equivalent center of gravity reference point that can characterize the trend of changes in the mass distribution of the human body. The center position of major body parts such as the head, torso, hips, and lower limbs can be determined first, and then different weights can be assigned according to the representativeness of each part in the human body to obtain the center of gravity monitoring information that moves continuously with the change of posture.

[0035] Preset abnormal relative parameter conditions can be established by pre-collecting normal activity samples and abnormal slip samples. For example, in scenarios such as elderly care rooms, beside nursing beds, and outside bathrooms, sequences of relative angle changes and relative distance changes of normal actions such as sitting down, lying down, bending over, picking up objects, and adjusting posture with support can be collected. Corresponding sequences of abnormal processes such as simulated flaccid slips, slow sliding down a wall after losing support, and weak slips from the edge of a bed can also be collected and processed. The range of change, sequence of change, and recovery characteristics of normal actions can be extracted, and the abrupt change characteristics, continuous shift characteristics, and non-recovery characteristics after imbalance of abnormal actions can be extracted to form preset abnormal relative parameter conditions. Preset abnormal relative parameter conditions can be stored in the form of rule sets. For example, when the relative angle between the trunk and lower limbs continuously changes in the direction of instability in a short period of time and does not return to the normal posture range after the change, or when the center of gravity shifts significantly relative to the center of the trunk, and at the same time, the distance distribution between the head, hips, and center of gravity shows abnormal rearrangement.

[0036] For example, by using the rate of change of relative angles and multidimensional relative distances based on the center of gravity, it is possible to deeply analyze the inherent posture distortions of the human body during a fall, without relying on the absolute strength of the overall signal, but focusing on the relative changes between different parts of the body. Therefore, even when the radar main beam deviates, resulting in a weak overall echo, it is possible to confirm body stability by identifying abnormal folds in the body structure or imbalances in the center of gravity, thus improving the accuracy of complex fall posture recognition. Furthermore, since the focus is on whether the structural relationships still conform to a controlled posture, even if the total point cloud is reduced, as long as key parts can still be partially tracked, instability can be identified from the relative relationships. This provides a strong ability to confirm abnormal postures even in situations such as antenna angle shifts, partial obstruction, and enhanced ground reflection.

[0037] In a flowchart of a life perception pattern recognition method based on millimeter-wave radar provided in one embodiment of this application, regarding the above step S130, the method may include, but is not limited to, step S310.

[0038] Step S310: When, within a preset abnormal stillness duration, the user's body height information obtained from monitoring points of multiple body parts is continuously lower than the preset height, and the vital signs monitoring information continuously fails to meet the preset typical vital signs range, and the movement distance of the user's body parts determined by monitoring points of multiple body parts is less than the preset movement distance, the abnormal stillness state of the user's body is determined, and an abnormal low-level stillness marker is triggered.

[0039] For example, within a pre-set preset abnormal stillness duration, multiple conditions can be continuously monitored, including the user's body height calculated from the monitoring point information being consistently lower than the preset height, the extracted vital signs monitoring information such as breathing or heart rate being consistently outside the preset typical vital signs range, and the movement distance of each body part monitoring point in three-dimensional space always being less than the preset movement distance; only when multiple conditions are simultaneously and continuously met throughout the entire time window is it finally determined that the user's body is in an abnormal stillness state and an abnormal low-level stillness marker is triggered.

[0040] To avoid the loss of applicability of specific values ​​under different installation heights, room layouts, and monitoring objects, the preset abnormal stillness duration, preset height, preset typical vital sign range, and preset movement distance can all be obtained using scenario-based calibration methods. Specifically, after the equipment is installed, the reference height of the ground, bed height, commonly used chair height, and wheelchair standing height in the room can be collected first to establish the normal activity height distribution in the current scenario; then, the body height information and body part movement characteristics of the target object or similar users in normal standing, sitting, lying down, bending over, short-term squatting, and tidying up items on the ground can be collected to form normal low-level activity samples; at the same time, the point cloud changes and vital sign changes in simulated abnormal low-level stillness can be collected to form abnormal samples; based on the relevant abnormal samples, the preset height is set to the judgment threshold of being lower than the normal sitting body height and close to the ground standing area, the preset movement distance is set to the judgment threshold that can exclude normal fine-tuning movements but retain the true stillness state, and the preset abnormal stillness duration is set to the duration threshold that is sufficient to filter out short-term low-level activities. For the preset typical vital signs range, it is not limited to a fixed numerical range. Instead, an individualized or scenario-based reference range is established based on indicators such as respiratory intensity, respiratory rhythm stability, and heartbeat extractability continuously collected by the device under normal conditions. When the vital signs monitoring information does not meet the reference range, it indicates that the vital signs are significantly weakened, the rhythm is abnormally disordered, or it is still difficult to stably extract effective vital signs after excluding obvious body movement interference.

[0041] Furthermore, the fact that the user's body part moves less than the preset movement distance can be determined by using multiple major body parts over a continuous period of time. This is because a single part may produce false displacement due to noise, clothing movement, or changes in local reflections. If the head, torso, hips, and lower limbs lack active displacement over a continuous period of time, it is more indicative that the target object is in an abnormally still state.

[0042] Based on this, in environments where strong background clutter makes instantaneous vital signs unreliable, a three-dimensional persistence strategy based on time windows—height, vital signs, and displacement—enhances the ability to identify true coma or weakness, ensuring that the marker is only triggered when the target is truly in a dangerous situation of prolonged, low position and lack of vitality. This minimizes the false alarm rate while ensuring early warning sensitivity and maintaining sensitivity to slow dangerous events.

[0043] In the aforementioned basic scheme, non-voluntary posture change markers are triggered by the difference between the movement trajectory and the preset normal posture change trajectory. However, in actual elderly care environments, the macroscopic trajectory of an elderly person slowly sitting down (e.g., sitting on a soft mattress or sofa) and the trajectory of someone slowly sliding down the edge of the bed to the ground due to exhaustion can be extremely similar in macroscopic radar point cloud representation. If only the difference in spatial trajectories is relied upon for comparison, misjudgment is easily made when facing soft supports, mistaking normal resting movements for dangerous slips. To solve this problem, it is necessary to further introduce interactive feedback information between the target object and the surrounding support environment, and to accurately distinguish between controlled sitting down and uncontrolled slips by analyzing the force deformation of the support environment.

[0044] like Figure 2 As shown, Figure 2 This is a flowchart of a life perception pattern recognition method based on millimeter-wave radar provided in another embodiment of this application. Regarding the above step S120, the method may include, but is not limited to, steps S410 to S420.

[0045] Step S410: Obtain support monitoring information of the support environment of the user's body based on millimeter-wave radar monitoring, obtain the support environment deformation index through the support monitoring information, and obtain the support environment feedback score between the user's body and the support environment of the user's body through the change rate of the support environment deformation index and the body monitoring information. Step S420: Analyze the user's body movement trajectory through body monitoring information. If the difference between the movement trajectory and the preset normal posture change trajectory is within the preset trajectory range, and the support environment feedback score exceeds the preset support environment threshold score, trigger the non-autonomous posture change marker.

[0046] For example, millimeter-wave radar can be positioned on the ceiling, side wall of a bed, or above a sofa, covering the human activity area and supporting areas such as the bed surface, cushions, edges, and armrests. By continuously transmitting detection signals and receiving echoes, after static background modeling, the main human point cloud is first separated. Then, background scattering points adjacent to the human contact point are searched within the vicinity of the main human point cloud as candidate point clouds for the supporting environment. Subsequently, the position changes, echo intensity changes, and phase changes of the candidate point clouds are tracked over multiple consecutive sampling times to obtain support monitoring information. This support monitoring information of the supporting environment refers to the minute displacements or reflection characteristic changes of background objects in contact with the human body, excluding the main human point cloud, as identified by the radar. For example, when a user slowly sits on a sofa, the echo in the corresponding area of ​​the sofa cushion shows a continuous downward trend, and the local reflection center shifts stably with the direction of force. When a user slides down the edge of the bed, the echo position of the rigid frame of the bed remains basically unchanged; only the human point cloud shifts downward. In this case, the corresponding area of ​​the supporting environment is unlikely to show a synchronous deformation response.

[0047] For example, when a human body comes into contact with a soft support, the support deforms, which manifests in radar echoes as slight subsidence or phase changes in local reflective surfaces. The corresponding support environment deformation index is used to quantify the degree and speed of deformation. The support environment deformation index is obtained by using support monitoring information. First, the candidate point cloud of the support environment can be divided into spatial regions, such as the bed surface center area, bed edge area, sofa seat area, armrest area, etc. Then, the continuous displacement trend, deformation duration, and whether the deformation response of each region is consistent with the moment of human contact are statistically analyzed over a period of time. If a region rapidly exhibits continuous, smooth, and directional displacement changes after the human body begins to sink, then the support environment deformation index of that region is high; if the region only shows scattered shaking or does not form a continuous change synchronized with human contact, then the support environment deformation index is low.

[0048] The rate of change of body monitoring information represents the speed and acceleration of the human body's descent. It can be obtained by tracking the height changes of key parts of the human body's main point cloud, such as the head, torso center, and pelvic region, over continuous time. This is combined with multi-frame smoothing to reduce the impact of clutter and occasional point jumps.

[0049] The support environment feedback score combines environmental deformation with human motion to assess whether the kinetic energy of a falling body is effectively absorbed and fed back by the support environment. The support environment feedback score is obtained by comparing the support environment deformation index with the rate of change of body monitoring information. This score can be constructed using a tiered approach. For example, when the falling body and environmental deformation show strong consistency in terms of start time, direction of change, duration, and magnitude of change, the support is considered effective, and the support environment feedback score is low. Conversely, when the body sinks significantly, but the environmental deformation is weak, delayed, or discontinuous, the support is considered insufficient, and the support environment feedback score is high.

[0050] The preset support environment threshold score can be obtained through sample learning or manual calibration. In one embodiment, typical action samples from multiple elderly care scenarios can be collected first, including sitting normally on the bed, sitting normally on the sofa, slowly sitting down while holding onto the edge of the bed, sliding down the edge of the bed, and falling due to instability while sitting. For each sample, it is marked whether it belongs to a controlled action and the corresponding support environment feedback score range is recorded. Subsequently, the score boundary that can reliably distinguish between sufficient and insufficient support is selected from the samples as the preset support environment threshold score.

[0051] For example, when a user changes from a standing to a sitting posture, if the radar continuously detects the gradual downward movement of the torso center and the point cloud of the sofa seat area shows continuous sinking within a similar timeframe, and the sinking process only gradually stops when the person is stably seated, then this can be identified as a state where human movement matches environmental deformation. If an elderly person slides down the edge of the bed, although the body is also sinking, the deformation of the bed edge is minimal, or the deformation is severely mismatched with the body's sinking speed. In this case, the calculated support environment feedback score will increase and exceed the preset support environment threshold score. For example, during the sliding down the bed edge, the torso and pelvic areas are detected to continuously sink, but the background point cloud of the bed edge area only shows slight disturbances and does not form a stable deformation trajectory synchronized with the body's sinking. In this case, it is determined that the human movement is not absorbed by the support environment.

[0052] Based on this, this application upgrades the simple human motion trajectory analysis to human-environment interaction analysis by supporting environmental feedback scoring. In environments with slight radar deviation or strong clutter, even if the human trajectory is blurred, as long as the mismatch between environmental deformation and human motion can be identified, the uncontrolled slip can be accurately identified, reducing the false alarm rate in high-frequency activity areas such as bedside or sofa side.

[0053] In another embodiment of this application, the life perception pattern recognition method based on millimeter-wave radar may include, but is not limited to, step S510, the above step S420.

[0054] Step S510: Obtain multiple kinematic features of the user's body through body monitoring information. If each kinematic feature meets the corresponding preset elastic range, the difference between the movement trajectory and the preset normal posture change trajectory is within the preset trajectory range, and the support environment feedback score exceeds the preset support environment threshold score, trigger the non-autonomous posture change marker.

[0055] For example, in millimeter-wave radar point cloud processing, the overall outline of the user object can first be obtained through human body point cloud clustering. Then, based on the vertical height distribution, horizontal expansion range, connectivity, and historical frame tracking results, the point cloud is divided into head, torso, upper limb, and lower limb regions. For each region, the center position, direction change, and local swing amplitude are continuously extracted to obtain the velocity change, acceleration change, and angle change trend of the corresponding part. If a single frame of point cloud is not stable enough, trajectory smoothing and missing point filling can be performed by combining multiple consecutive frames to improve the stability of kinematic features. Kinematic features may include, but are not limited to, the instantaneous velocity, acceleration, angular velocity, and fluctuation frequency of various major parts of the body over a short period of time.

[0056] A preset elastic range refers to a specific range of kinematic parameters caused by muscle relaxation or weakness during involuntary falls. Preset elastic ranges can be established using an offline sample database. This involves collecting samples from various scenarios, including normal sitting, active squatting, squatting while holding onto something, accidental falls, falls due to fainting, and falls from the bedside. The kinematic characteristics of the head, trunk, and limbs in each sample are then statistically analyzed. Feature combinations representing loss of control are selected and recorded as preset elastic ranges for different action scenarios. For example, in normal controlled movements, the trunk often actively decelerates, pauses briefly, or makes minor adjustments in the opposite direction near the end of its descent. In involuntary falls, the trunk speed may continuously change in one direction without buffering, and the limbs lack the rapid, coordinated movements required before active support. By summarizing these differences, a preset elastic range for judgment can be formed.

[0057] For example, multiple kinematic features can be divided into several judgment items, such as whether there is significant deceleration at the end of the trunk descent, whether the adjustment of the head and trunk maintains active posture correction, whether the upper limbs provide effective support, and whether the forward swing of the lower limbs provides coordinated flexion and extension buffering. If the extraction results in the judgment items all point to the loss of control mode, then each kinematic feature can be considered to meet the corresponding preset elastic range.

[0058] It is understood that the embodiments of this application combine macroscopic trajectory analysis with microscopic kinematic feature analysis. Trajectory analysis can identify whether a significant posture change has occurred, while kinematic feature analysis can identify whether the posture change involves active control and buffering. Thus, in scenarios with hard surfaces or where environmental feedback is not obvious, by identifying kinematic feature changes caused by muscle loss of control, the shortcomings of environmental deformation analysis can be compensated for. This allows the monitoring mechanism to maintain high sensitivity in identifying involuntary slippage in various complex physical environments, further improving the comprehensiveness and accuracy of the early warning.

[0059] In another embodiment of this application, a life perception pattern recognition method based on millimeter-wave radar is provided, in which a preset normal posture change trajectory is obtained from the personalized posture change trajectory library of the corresponding user object by means of the type of motion trajectory and the timestamp.

[0060] For example, the type of motion trajectory refers to classifying the current action into a specific basic action category through preliminary morphological analysis. This can be achieved by initially identifying the direction, duration, starting posture, ending posture, and main trunk motion trend of the current point cloud trajectory. The personalized posture change trajectory library is a collection of exclusive action templates that are continuously collected, learned, and updated for specific user groups. It can distinguish different action types and further classify the same type of action according to timestamps.

[0061] For example, by continuously collecting normal action data of users in their daily lives, and combining the results of manual review, nursing records, or user self-confirmation, the samples are labeled with tags such as normal sitting down, normal standing up, normal bending over, and adjusting posture at the bedside at night. Then, the samples are classified and stored according to user identity, action type, and time period. For each type of sample, representative trajectory segments, common duration ranges, displacement trends of major body parts, and stable posture characteristics when the action is completed are retained, thereby establishing a personalized posture change trajectory library for the corresponding user.

[0062] For example, if the elderly person's trajectory of getting up from the bedside around 7 AM is recorded as slow with slight pauses, while their trajectory of sitting on the living room sofa around 3 PM is relatively continuous, this data can be categorized and stored in a personalized posture change trajectory library. When a slow descent trajectory is detected at 10 PM, the type of movement and timestamp are extracted. Then, a preset normal posture change trajectory for sitting down at night is retrieved from the elderly person's trajectory library for comparison. The overall descent trend, trunk movement sequence, movement duration rhythm, and the stabilization pattern at the end of the movement are compared. For instance, if the user typically holds onto the edge of the bed before slowly descent and pauses briefly before touching the bed, and the current trajectory lacks this pause and shows continuous sliding, then even if the overall trajectory is still a slow descent, it can be identified as deviating from the user's normal pattern for that time period. In this way, by matching personalized trajectory libraries and timestamps, accurate customized monitoring can be achieved, identifying subtle deviations between the current movement and the user's daily habits, thereby improving the ability to identify atypical dangerous events without increasing false alarms.

[0063] In another embodiment of this application, the life perception pattern recognition method based on millimeter-wave radar may include, but is not limited to, steps S610 to S630, regarding step S510.

[0064] Step S610: Obtain the relative motion coordination parameters between different body parts through the monitoring point information of different body parts; obtain the support stability parameters between the user's body and the support environment through the support monitoring information and the monitoring point information of different body parts; obtain the stability parameters through the support stability parameters and the relative motion coordination parameters. Step S620: Obtain physiological micro-motion continuity parameters through vital sign monitoring information; Step S630: When all kinematic features meet the corresponding preset elasticity range, the difference between the motion trajectory and the preset normal posture change trajectory is within the preset trajectory range, the support environment feedback score exceeds the preset support environment threshold score, the stability parameter is less than the preset stability threshold, and the physiological micro-motion continuity parameter exceeds the preset physiological continuity threshold, trigger the non-voluntary posture change marker, or... A kinematic score is obtained based on the number of kinematic features within the corresponding preset elastic range; a trajectory score is obtained based on the difference between the motion trajectory and the preset normal posture change trajectory; a stability score is obtained based on the difference between the stability parameter and the preset stability threshold; and a physiological continuity score is obtained based on the difference between the physiological micromotion continuity parameter and the preset physiological continuity threshold. A comprehensive posture score is obtained based on the kinematic score, trajectory score, stability score, and physiological continuity score. If the comprehensive posture score is less than the preset abnormal posture score threshold, an involuntary posture change marker is triggered.

[0065] Understandably, in extremely complex scenarios, an elderly person's fall may not be a standard case of loss of control. For example, in the initial stage of a fall, the elderly person may still retain some consciousness and attempt to grab onto surrounding objects to save themselves, resulting in a complex state of semi-uncontrolled, semi-struggling movements of various parts of the body. In such cases, relying solely on logic and conditions to trigger a flag could easily lead to the entire dangerous event being missed if a particular local feature fails to meet the conditions.

[0066] Therefore, in some embodiments, the relative motion coordination parameters between different body parts are first obtained through monitoring point information of different body parts. The relative motion coordination parameters are used to measure the degree of coordination between different body parts during movement. The monitoring point information of different body parts can be set as head monitoring point, shoulder monitoring point, trunk center monitoring point, pelvis monitoring point, upper limb end monitoring point, and lower limb end monitoring point. The motion trajectory is output through the position and direction change information of the monitoring points in continuous time. Then, the relative motion coordination parameters are obtained according to whether they are started synchronously, whether they are transmitted continuously along a reasonable human joint chain, and whether there is a sudden mismatch. For example, when sitting normally, the pelvis moves down first, the trunk adjusts accordingly, the knees and ankles flex and extend in coordination, and the upper limbs may provide assistance, presenting a coherent and coordinated overall movement. However, in an uncontrolled or semi-uncontrolled slip, the trunk may lose balance quickly first, while the lower limbs fail to adjust synchronously, or one upper limb may swing suddenly while the other remains stiff, resulting in a decrease in the relative motion coordination parameters.

[0067] By analyzing support monitoring information and monitoring point information from different body parts, the support stability parameters between the user's body and the support environment are obtained. These parameters characterize the force distribution and stability at the contact points between the body and the external environment. They can be indirectly inferred from support monitoring information and monitoring point information from different body parts. Specifically, the system can identify the contact relationship between the human body and locations such as the bed edge, bed surface, sofa cushion, floor, and armrests, and track the continuity of contact points during movement, the stability of the contact position, and whether the body's center of gravity projection always falls near the effective support area. If the contact area is stable, the support position is not suddenly lost, and the body's main mass area does not leave the support boundary during the sinking process, the support stability parameters are high. Conversely, if contact points frequently switch, the support area suddenly decreases, and the center of gravity significantly deviates from the support boundary, the support stability parameters are low.

[0068] For example, physiological micro-motion continuity parameters can be obtained through vital sign monitoring information. These parameters can be acquired by analyzing the continuous changes in minute vital signs such as respiration and heart rate during movement. For instance, vital sign monitoring information can be directly obtained by detecting chest and abdominal micro-movements using millimeter-wave radar under resting or low-speed movement conditions, or it can be provided by a bedside vital sign monitoring module, wearable device, or seat pressure micro-motion sensor linked to the monitoring system. In one embodiment, millimeter-wave radar can be used to extract minute reciprocating displacements from the chest region to form a respiratory-related micro-motion sequence. If the user is wearing a wristband-type vital sign device, heart rate rhythm changes can be further received as auxiliary information. Subsequently, the continuity of respiratory micro-movements before, during, and after the movement is analyzed to determine whether the rhythm is interrupted, whether there are abrupt changes, or whether there are short pauses or abnormal rapid fluctuations, thereby obtaining physiological micro-motion continuity parameters.

[0069] For example, the non-voluntary posture change marker has two triggering mechanisms, including a serial logic judgment, which requires that kinematic features, trajectory differences, support environment feedback, stability parameters, and physiological micro-motion continuity parameters all meet their respective abnormal threshold conditions, and is suitable for typical completely uncontrolled slip scenarios with clear features; it also includes a more flexible comprehensive scoring mechanism, which converts the features of each dimension into specific scores and performs weighted fusion to obtain a comprehensive posture score.

[0070] In one embodiment, the comprehensive scoring mechanism can adopt a tiered scoring method. When there are many kinematic features falling within a preset elastic range, the kinematic score is tilted towards the abnormal side; when the current trajectory deviates significantly from the preset normal posture change trajectory, the trajectory score is tilted towards the abnormal side; when the stability parameter is significantly lower than the preset stability threshold, the stability score is tilted towards the abnormal side; when the physiological micromotion continuity parameter significantly exceeds the preset physiological continuity threshold, the physiological continuity score is tilted towards the abnormal side; subsequently, corresponding weights are set according to the importance of each score to form a comprehensive posture score.

[0071] For example, in a semi-uncontrolled slip incident, an elderly person's torso slowly sinks, but their right hand attempts to grab the edge of the bed. At this time, the active movement of the right hand may cause some kinematic features to not fall completely within the preset elastic range. If a serial logic judgment is used, it may be missed because local features do not meet the conditions. However, using a comprehensive scoring mechanism, a lower trajectory score, a very low stability score, and a high physiological continuity score will be calculated. Therefore, even if the kinematic score is at a critical state, the comprehensive posture score calculated by combining the four scores will still be lower than the preset abnormal posture score threshold, thus successfully triggering the non-voluntary posture change marker.

[0072] Based on this, the embodiments of this application, through in-depth analysis of coordination, stability and physiological micro-movements, combined with a comprehensive scoring judgment mechanism, can tolerate the absence or deviation of individual features, rely on the overall synergy of multi-dimensional data to approximate the truth, improve the fault tolerance rate in extremely complex scenarios, and avoid missed detections caused by local action interference.

[0073] like Figure 3 As shown, Figure 3 This is a flowchart of a life perception pattern recognition method based on millimeter-wave radar provided in another embodiment of this application. Regarding the above step S110, the method may include, but is not limited to, steps S710 to S740.

[0074] Step S710: Obtain the walking aid monitoring information and ground monitoring information obtained based on millimeter-wave radar monitoring, as well as the initial monitoring information of the user's body; Step S720: Obtain the contact vibration characteristics of the walking aid through the walking aid monitoring information, and obtain the ground deformation characteristics through the ground monitoring information; Step S730: Based on the periodicity of ground deformation characteristics, obtain gait rhythm information; based on the contact vibration characteristics of the walking aid and ground deformation characteristics, obtain ground landing event information and ground takeoff event information. Step S740: Update the initial monitoring information of the lower body in the initial monitoring information using gait rhythm information, landing event information and take-off event information to obtain body monitoring information.

[0075] It is understandable that when elderly people use walking aids, such as canes, four-legged crutches, or walking frames, for daily activities, the metal or rigid structure of the walking aid will generate radar echoes that overlap with the echoes from the lower body of the elderly person. This can lead to difficulties in extracting information from the lower body monitoring points, broken tracks, or drift.

[0076] For example, it is possible to obtain walking aid monitoring information and ground monitoring information based on millimeter-wave radar monitoring, as well as initial monitoring information of the user's body. The initial monitoring information refers to the human body point cloud data obtained by radar preliminary scanning and clustering. In the presence of a walking aid, the lower body part in the initial monitoring information includes the interference points of the walking aid. The walking aid monitoring information and ground monitoring information are feature data extracted from the walking aid itself and the ground reflection area.

[0077] When the walker touches the ground or the foot steps, a tiny Doppler frequency shift or phase oscillation occurs at the moment of contact. By identifying the transient changes, the contact vibration characteristics of the walker can be obtained. At the same time, when the foot or walker is pressed against the ground, it will cause a tiny displacement of the local reflective surface. By tracking the tiny displacement, the ground deformation characteristics can be obtained.

[0078] Furthermore, since normal walking is periodic, gait rhythm information representing walking speed and stride frequency can be extracted based on the periodic undulations of ground deformation features. By aligning and fusing the walking aid's contact vibration features with ground deformation features on the time axis, the exact moment of each step or walking aid contacting and leaving the ground can be located, thus obtaining ground contact event information and ground departure event information. Therefore, after acquiring rhythm and event information, instead of simply relying on cluttered point clouds for clustering, gait rhythm information, ground contact event information, and ground departure event information can be used as a time reference to perform temporal separation, filtering, and correction of the initial lower body monitoring information in the initial monitoring data, eliminating interference points that do not conform to gait rhythm, and obtaining accurate body monitoring information.

[0079] It is understood that the embodiments of this application transform spatial point cloud separation into multi-dimensional feature fusion based on temporal rhythms and physical contact events. Even if the walking aid and lower limbs completely overlap in space, the actual movement trajectory of the human body can be accurately separated through ground vibration and gait cycle, thereby providing highly reliable data for subsequent posture change and stability analysis, overcoming the interference of strong echoes from the walking aid on the lower body point cloud, and improving the accuracy of monitoring elderly people slipping while using walking aids.

[0080] In another embodiment of this application, the life perception pattern recognition method based on millimeter-wave radar may include, but is not limited to, steps S810 to S820, regarding step S740.

[0081] Step S810: Using the initial upper body monitoring information and gait rhythm information in the initial monitoring information, obtain the predicted search area for the initial lower body monitoring information, and the predicted lower body monitoring information in the predicted search area. Step S820: Based on the lower body prediction monitoring information, update the initial lower body monitoring information obtained in the prediction search area to obtain the predicted optimized lower body monitoring information. Furthermore, through gait rhythm information, landing event information, and take-off event information, interpolate and optimize the predicted optimized lower body monitoring information to obtain body monitoring information.

[0082] For example, during human walking, there is an inherent biomechanical linkage between the movement of the upper and lower body. Since the upper body is less affected by the walking aid, the initial monitoring information is relatively clear and stable. Therefore, by analyzing the initial monitoring information of the upper body and combining it with the acquired gait rhythm information, the spatial range in which the lower body should appear at the next moment can be calculated using the human kinematic model, i.e., the predicted search area. At the same time, the expected posture and position of the lower body in this area can be estimated, i.e., the lower body predicted monitoring information.

[0083] Subsequently, the predicted search area can be used as a spatial filter, matching and updating the initial lower body monitoring information acquired only within this area, eliminating free noise and assistive device interference points outside the area, thus obtaining preliminary predicted and optimized lower body monitoring information. However, in actual walking, the lower limbs may be completely obscured by the assistive device, resulting in the complete loss of point cloud data at certain moments. In this case, gait rhythm information, landing event information, and take-off event information can be used to determine which stage of the stride cycle is currently in. Based on the determined stage node, the missing trajectory segments in the predicted and optimized lower body monitoring information are interpolated and optimized in accordance with human movement patterns, smoothly connecting the broken trajectories to obtain complete and continuous body monitoring information. This enhances the trajectory reconstruction capability in occlusion and aliasing environments, ensuring that even under poor radar signal quality, a smooth, continuous, and logically consistent lower body movement trajectory can be output.

[0084] Secondly, embodiments of this application provide a life-sensing pattern recognition system based on millimeter-wave radar, comprising: The information acquisition module is used to acquire body monitoring information of the user's body obtained based on millimeter-wave radar monitoring. The body monitoring information includes monitoring point information of multiple body parts of the user and vital sign monitoring information. The abnormal trajectory module is used to analyze the user's body movement trajectory through body monitoring information when it is determined that the user's body has a slow and continuous downward trend through monitoring information of multiple body parts. If the difference between the movement trajectory and the preset normal posture change trajectory is within the preset trajectory range, a non-voluntary posture change marker is triggered. The abnormal state module is used to obtain relative parameters between different body parts based on non-autonomous posture change markers and monitoring point information of different body parts. When the relative parameters meet the preset abnormal relative parameter conditions, an abnormal stability change marker is triggered. When the abnormal body position of the user is determined to be abnormal through monitoring point information of multiple body parts, and the abnormal static state of the user's body is determined through vital sign monitoring information and monitoring point information of multiple body parts, an abnormal low-position static marker is triggered. The early warning module is used to trigger warnings of dangerous events based on abnormal stability change markers and abnormal low-level stationary markers.

[0085] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A life-sensing pattern recognition method based on millimeter-wave radar, characterized in that, include: Obtain body monitoring information of a user's body based on millimeter-wave radar monitoring. The body monitoring information includes monitoring point information of multiple body parts of the user and vital sign monitoring information. When it is determined through monitoring information from multiple body parts that the user's body has a slow and continuous downward trend, the movement trajectory of the user's body is analyzed through the body monitoring information. If the difference between the movement trajectory and the preset normal posture change trajectory is within the preset trajectory range, a non-voluntary posture change marker is triggered. Based on the non-autonomous posture change marker, relative parameters between different body parts are obtained through monitoring point information of different body parts. When the relative parameters meet the preset abnormal relative parameter conditions, an abnormal stability change marker is triggered. When the abnormal body position of the user is determined through the monitoring point information of the multiple body parts, and the abnormal static state of the user's body is determined through the vital sign monitoring information and the monitoring point information of the multiple body parts, an abnormal low-position static marker is triggered. Based on the aforementioned abnormal stability change markers and abnormal low-level stationary markers, a danger event warning is triggered.

2. The life perception pattern recognition method based on millimeter-wave radar according to claim 1, characterized in that, The relative parameters include relative angle and relative distance; The process involves obtaining relative parameters between different body parts through monitoring point information from different body parts. If these relative parameters meet preset abnormal relative parameter conditions, an abnormal stability change flag is triggered, including: By monitoring information from different body parts, the relative angles and distances between different body parts, as well as the center of gravity monitoring information, can be obtained. An abnormal stability change flag is triggered when the rate of change of the relative angle meets the corresponding preset abnormal relative parameter conditions; or, when the relative distance between the center of gravity monitoring information and the monitoring point information of different body parts, as well as the relative distance between the monitoring point information of different body parts, both meet the corresponding preset abnormal relative parameter conditions.

3. The life-sensing pattern recognition method based on millimeter-wave radar according to claim 1, characterized in that, When an abnormal user's body position is determined through monitoring point information from multiple body parts, and an abnormal static state of the user's body is determined through vital sign monitoring information and monitoring point information from multiple body parts, triggering an abnormal low-level static marker includes: If, within a preset abnormal stillness period, the user's body height information obtained from monitoring points of multiple body parts is consistently lower than a preset height, and the vital signs monitoring information consistently fails to meet the preset typical vital signs range, and the movement distance of the user's body parts determined by monitoring points of multiple body parts is less than a preset movement distance, an abnormal stillness state of the user's body is determined, and an abnormal low-level stillness marker is triggered.

4. The life perception pattern recognition method based on millimeter-wave radar according to claim 1, characterized in that, The step of analyzing the user's body movement trajectory through the body monitoring information, and triggering a non-voluntary posture change marker when the difference between the movement trajectory and the preset normal posture change trajectory is within the preset trajectory range, includes: The system acquires support monitoring information of the support environment of the user's body based on millimeter-wave radar monitoring, obtains a support environment deformation index through the support monitoring information, and obtains a support environment feedback score between the user's body and the support environment of the user's body through the rate of change of the support environment deformation index and the body monitoring information. The user's body movement trajectory is analyzed by the body monitoring information. If the difference between the movement trajectory and the preset normal posture change trajectory is within the preset trajectory range, and the support environment feedback score exceeds the preset support environment threshold score, a non-autonomous posture change marker is triggered.

5. The life-sensing pattern recognition method based on millimeter-wave radar according to claim 4, characterized in that, The triggering of a non-autonomous posture change marker when the difference between the motion trajectory and the preset normal posture change trajectory is within the preset trajectory range, and the support environment feedback score exceeds the preset support environment threshold score, includes: The user's body is obtained through the body monitoring information. When each kinematic feature meets the corresponding preset elasticity range, the difference between the movement trajectory and the preset normal posture change trajectory is within the preset trajectory range, and the support environment feedback score exceeds the preset support environment threshold score, a non-autonomous posture change marker is triggered.

6. The life-sensing pattern recognition method based on millimeter-wave radar according to claim 5, characterized in that, The preset normal posture change trajectory is obtained from the corresponding user object's personalized posture change trajectory library by using the type and timestamp of the motion trajectory.

7. The life-sensing pattern recognition method based on millimeter-wave radar according to claim 6, characterized in that, The non-autonomous posture change flag is triggered when all kinematic features meet the corresponding preset elasticity range, the difference between the motion trajectory and the preset normal posture change trajectory is within the preset trajectory range, and the support environment feedback score exceeds the preset support environment threshold score. This includes: The relative motion coordination parameters between different body parts are obtained by monitoring point information of different body parts. The support stability parameters between the user's body and the support environment are obtained by the support monitoring information and the monitoring point information of different body parts. The stability parameters are obtained by the support stability parameters and the relative motion coordination parameters. Physiological micro-motion continuity parameters are obtained through the vital sign monitoring information; When all kinematic features meet the corresponding preset elasticity range, the difference between the motion trajectory and the preset normal posture change trajectory is within the preset trajectory range, the support environment feedback score exceeds the preset support environment threshold score, the stability parameter is less than the preset stability threshold, and the physiological micro-motion continuity parameter exceeds the preset physiological continuity threshold, a non-voluntary posture change marker is triggered, or... A kinematic score is obtained based on the number of kinematic features within the corresponding preset elastic range; a trajectory score is obtained based on the difference between the motion trajectory and the preset normal posture change trajectory; a stability score is obtained based on the difference between the stability parameter and the preset stability threshold; and a physiological continuity score is obtained based on the difference between the physiological micromotion continuity parameter and the preset physiological continuity threshold. A comprehensive posture score is obtained based on the kinematic score, trajectory score, stability score, and physiological continuity score. If the comprehensive posture score is less than the preset abnormal posture score threshold, a non-voluntary posture change marker is triggered.

8. The life perception pattern recognition method based on millimeter-wave radar according to claim 1, characterized in that, The acquisition of body monitoring information of the user's body obtained based on millimeter-wave radar monitoring includes: Acquire walking aid monitoring information and ground monitoring information based on millimeter-wave radar monitoring, as well as initial monitoring information of the user's body; The walking aid's contact vibration characteristics are obtained through the walking aid's monitoring information, and the ground deformation characteristics are obtained through the ground monitoring information. Based on the periodicity of the ground deformation characteristics, gait rhythm information is obtained; based on the walking aid contact vibration characteristics and ground deformation characteristics, ground landing event information and ground takeoff event information are obtained. The initial lower body monitoring information in the initial monitoring information is updated using the gait rhythm information, landing event information, and takeoff event information to obtain body monitoring information.

9. The life-sensing pattern recognition method based on millimeter-wave radar according to claim 8, characterized in that, The initial lower body monitoring information in the initial monitoring information is updated using the gait rhythm information, landing event information, and takeoff event information to obtain body monitoring information, including: Using the initial monitoring information of the upper body and gait rhythm information in the initial monitoring information, a predicted search area for the initial monitoring information of the lower body is obtained, as well as the predicted monitoring information of the lower body in the predicted search area. Based on the lower body prediction monitoring information, the initial lower body monitoring information obtained in the prediction search area is updated to obtain the prediction optimized lower body monitoring information. Furthermore, the prediction optimized lower body monitoring information is interpolated and optimized using gait rhythm information, landing event information, and take-off event information to obtain body monitoring information.

10. A life-sensing pattern recognition system based on millimeter-wave radar, characterized in that, include: The information acquisition module is used to acquire body monitoring information of the user's body obtained based on millimeter-wave radar monitoring. The body monitoring information includes monitoring point information of multiple body parts of the user and vital sign monitoring information. The abnormal trajectory module is used to analyze the user's body movement trajectory through the body monitoring information when it is determined that the user's body has a slow and continuous downward trend through the monitoring information of multiple body parts. If the difference between the movement trajectory and the preset normal posture change trajectory is within the preset trajectory range, a non-autonomous posture change marker is triggered. The abnormal state module is used to obtain relative parameters between different body parts based on the non-autonomous posture change marker through monitoring point information of different body parts. When the relative parameters meet the preset abnormal relative parameter conditions, it triggers the abnormal stability change marker. When it is determined that the user's body position is abnormal through the monitoring point information of multiple body parts, and the abnormal static state of the user's body is determined through vital sign monitoring information and monitoring point information of multiple body parts, it triggers the abnormal low-position static marker. The early warning module is used to trigger a warning of dangerous events based on the abnormal stability change marker and the abnormal low-level stationary marker.

Images