Bed fall detection method, device and medium for home health monitoring radar

Abstract

This invention relates to the field of radar signal processing technology, specifically to a bed-fall detection method, device, and medium for home health monitoring radars. This method deeply integrates one-dimensional range profile (HRRP) features (such as RCS amplitude) with point cloud information. By designing efficient feature mining methods and intelligent decision logic, it aims to effectively overcome the aforementioned challenges and achieve high-accuracy, low-false-alarm bed-fall detection in complex home environments. Specifically, it simultaneously utilizes two features: the number of point cloud clusters and the time series variation of radar cross-sectional area (RCS) amplitude to determine bed-fall. The number of point clouds reflects the spatial dispersion of the target, while the RCS amplitude reflects the target's reflection strength (a person falling may move away from the radar or their posture may change, causing a sudden drop in RCS). This multi-dimensional information fusion significantly improves the accuracy and robustness of bed-fall detection compared to single-feature detection, reducing false positives caused by environmental interference or local target motion.

Description

Technical Field

[0001] The present invention relates to the field of radar signal processing technology, and in particular to a method, device and medium for detecting falling out of bed for a home health monitoring radar. Background Art

[0002] Bed fall detection detects whether a caregiver has fallen from bed. This technology can promptly detect and issue alerts for abnormal out-of-bed behavior for specific individuals (such as those bedridden, those with limited mobility, those living in home care, and patients), ensuring their safety. It is a key health monitoring function, with its core goal being to promptly and accurately detect unexpected, potentially harmful falls from bed for individuals such as bedridden patients, elderly individuals with limited mobility, patients recovering from surgery, and elderly individuals living in home care. Prompt alerts for such incidents are crucial for ensuring the safety of these individuals, saving valuable rescue time, reducing the burden on caregivers, and improving the quality of home care. This represents a key development direction for intelligent health monitoring.

[0003] At present, the technical solutions for realizing fall-out detection mainly include the following categories:

[0004] Solutions based on physical contact / pressure sensors: For example, pressure sensor arrays or piezoelectric films are placed under the mattress or near the bed. These trigger an alarm when a person leaves a specific area of ​​the bed or exerts pressure on the bedside floor. The main limitations of this type of solution are:

[0005] Installation complexity and intrusiveness: Sensors need to be laid on the bed or floor, which may affect user comfort and the aesthetics of the home environment.

[0006] High false alarm rate: Easily affected by factors such as pets getting on the bed, heavy objects being placed on the bed, and users engaging in normal activities at the bedside (such as sitting on the bed and putting on shoes).

[0007] Limited coverage: Usually only a specific area (the bed surface or a small area of ​​the ground next to the bed) can be detected, and falls that are slightly further away may be missed.

[0008] Unable to distinguish between users and objects: It only senses pressure changes and cannot distinguish whether the user has fallen out of bed or another object has fallen.

[0009] Solutions based on visual cameras (RGB / RGB-D): Utilize cameras to capture user images or depth information and use gesture recognition, behavior analysis, or area intrusion detection to determine if a person has fallen out of bed. The core challenges of this type of solution are:

[0010] Risk of privacy leakage: Continuous video surveillance involves a high degree of user privacy, which is particularly sensitive in a home environment. User acceptance is low and there are data security risks.

[0011] Poor environmental adaptability: It is very sensitive to lighting conditions (night, strong light), obstructions (quilts, mosquito nets, furniture), user clothing (dark clothing), etc., and lacks robustness.

[0012] High computational complexity: Real-time, accurate posture recognition or behavior analysis requires high computing power, which increases system cost and power consumption.

[0013] Blind spot problem: The camera's viewing angle is fixed, and there are blind spots in monitoring (such as the other side of the bed).

[0014] Solutions based on wearable devices (such as wristbands and belts): Detect sudden falls or abnormal posture changes of users through accelerometers and gyroscopes. Their main disadvantages include:

[0015] User compliance issues: Users are required to wear the device continuously, which is not suitable for users with cognitive impairments or resistance to wearing the device. There is also the problem of forgetting to charge or feeling uncomfortable while wearing the device.

[0016] Misjudgment and missed judgment: Normal behaviors such as violent turning over and sitting up quickly may be misjudged as falling out of bed; while falling out of bed in a slow manner or in a specific posture may be missed because the acceleration characteristics are not obvious.

[0017] Device Dependency: Functionality is completely dependent on the user wearing and functioning of the device.

[0018] In recent years, millimeter-wave radar technology has shown great potential in the field of home health monitoring due to its unique advantages (non-contact, privacy-protecting, unaffected by light, able to penetrate thin obstructions, and capable of providing multi-dimensional information such as distance, speed, and angle), and is considered an ideal choice to address the pain points of the above-mentioned traditional solutions. However, applying millimeter-wave radar technology to high-reliability fall detection, especially in complex home environments, still faces significant technical challenges and development bottlenecks:

[0019] Difficulty in target representation and feature extraction: A person's posture in bed (lying flat, sideways, sitting up) and their posture on the ground after a fall (curled up, lying flat, sideways) may not have distinct and universally distinguishable features in radar echoes (especially one-dimensional range profiles or point clouds). The human body itself is a complex distributed target, and its radar cross-section (RCS) varies significantly with posture, orientation, and relative position to the radar.

[0020] Environmental clutter interference: Home environments often contain static and quasi-static clutter, such as bed frames, furniture, and other debris, as well as dynamic clutter, such as swaying curtains and running fans. These interferences can drown out weak signals of falling out of bed or create false targets, interfering with the accurate detection and tracking of real human targets.

[0021] Complexity of point cloud processing: While point cloud data is rich in information, it is also highly complex to process. Key challenges include efficiently and robustly clustering sparse, noisy point clouds to accurately distinguish a single human target (a person lying in bed) from multiple suspected targets (fragmented point clouds) due to posture changes after a fall or environmental reflections, or to distinguish between humans and distracting objects such as pets. Existing methods are prone to failure when processing point clouds with low signal-to-noise ratios or complex scenes.

[0022] Diversity of Behavior Patterns: Falling out of bed is inherently diverse (e.g., sudden falls, slow slides), and requires precise differentiation from normal getting out of bed and bedside activities (e.g., sitting up, standing up, and walking around the bed). Designing robust features and decision logic that can both highly sensitively capture abnormal falls and effectively suppress false alarms related to normal activities remains a key and challenging area of ​​current research.

[0023] Real-time and robustness requirements: As a safety feature, bed fall detection places extremely high demands on the system's real-time performance (rapid alarm) and long-term robustness (stable operation in various home environments and user states). Existing millimeter-wave radar-based solutions still need further improvement in these aspects.

[0024] In summary, while millimeter-wave radar offers a promising technology for home fall detection, the field remains in its exploratory stages. Existing solutions have significant shortcomings in detection accuracy in complex scenarios, robustness (especially in suppressing false alarms for normal waking behaviors), and efficient point cloud / signal feature mining, presenting significant research potential and room for technological optimization. Summary of the Invention

[0025] In view of this, the present invention proposes a bed fall detection device, device and medium for home health monitoring radar, which can achieve high-accuracy and low-false-alarm bed fall event detection in complex home environments.

[0026] To achieve the above object, the technical solution of the present invention is:

[0027] A method for detecting bed falls for a home health monitoring radar, including: performing one-dimensional range imaging and point cloud imaging processing on the original echo data collected by the radar to obtain the one-dimensional range image and point cloud information of the observation scene; performing clutter suppression on the one-dimensional range image, calculating the amplitude of the clutter-removed range spectrum data and performing peak search to find the point with the largest amplitude, using the value of this point as the amplitude of the target, obtaining the RCS amplitude time series of the observation scene, mining the features of the amplitude time series RCS, and counting the continuous frame number AccFlag1 below the threshold Th1 and the continuous frame number AccFlag2 above the threshold Th1; performing clustering on the obtained point cloud information to obtain the number ClusterNum of classes after point cloud clustering; based on the RCS features and the number ClusterNum of classes after point cloud clustering, determining abnormal bed fall behaviors and judging whether a bed fall event occurs.

[0028] Among them, the determination of abnormal bed fall behaviors is specifically implemented as follows:

[0029] Perform dynamic counting on the frames with RCS values below and above the threshold Th1, where AccFlag1 represents the number of frames with RCS values below Th1; AccFlag2 represents the number of frames with RCS values above Th1;

[0030] The specific implementation method of dynamic counting is:

[0031] 1) If the RCS value of the current frame is less than the threshold Th1, then increment AccFlag1 by 1, and at the same time check AccFlag2. If it is greater than 0, then decrement AccFlag2 by 1, otherwise clear AccFlag2 to 0;

[0032] 2) If the RCS value of the current frame is greater than the threshold Th1, then increment AccFlag2 by 1, and at the same time check AccFlag1. If it is greater than 0, then decrement AccFlag1 by 1, otherwise clear AccFlag1 to 0;

[0033] Design logical judgment conditions for bed fall detection and false alarm suppression. If AccFlag1>Th2 and AccFlag2>Th2, it is considered that a suspected bed fall behavior occurs. On this basis, check the number ClusterNum of classes after point cloud clustering of the current frame. If ClusterNum<Th3, then determine that a bed fall behavior occurs and output a bed fall flag; otherwise, it is considered normal activity beside the bed or inside the house.

[0034] Among them, the specific method of point cloud clustering is as follows:

[0035] Step1: Assume that the obtained point cloud set has a total of K points. Take the first point as the reference point, and assume that the coordinates of this point are ,in: is the signal-to-noise ratio, For distance, is the azimuth, is the pitch angle, is the Doppler velocity; and the label of each point is initialized as: GroupNo = [1, zeros(1,K - 1)], where the first point is the reference point and its label is 1, and the labels of the remaining K - 1 points are 0;

[0036] Step 2: Calculate the reference points separately Divide by point cloud collection Other points The Euclidean distance is as follows:

[0037]

[0038] Where: represents the absolute value, is a vector containing five-dimensional information: distance, azimuth, pitch angle, and Doppler velocity, where Represents the point cloud distance difference; Represents the point cloud azimuth difference; Represents the pitch angle difference of the point cloud; Represents the Doppler velocity difference of the point cloud; and The five-dimensional information difference of distance, azimuth, pitch angle and Doppler speed is used to make a threshold decision. If the decision condition is met, Points associated with reference points , set the label of the point to 1;

[0039] The judgment condition is as follows: the point cloud distance difference is less than the point cloud distance difference threshold, the point cloud azimuth difference is less than the point cloud azimuth difference threshold, the point cloud pitch angle difference is less than the point cloud pitch angle difference threshold, and the Doppler velocity difference is less than the point cloud Doppler velocity difference threshold;

[0040] Step 3: Find all points with label 1 in the label set GroupNo and count the number of these points N; if the number of points N is greater than the minimum point cloud number threshold TargetNum, the point set with label 1 is considered to be a valid target point set, and the number of classes ClusterNum is increased by 1;

[0041] Step 4: Collect in point cloud In the process, remove the points that have been associated with the same class, extract the set of remaining unassigned points, repeat the above steps until all points are condensed and classified, and obtain the number of condensed classes ClusterNum.

[0042] Among them, the clutter map method is used to suppress clutter on the one-dimensional range image of the observation scene.

[0043] Among them, the echo sampling point of the first Chirp signal in each frame is taken for fast Fourier transform to obtain the one-dimensional range image of the observation scene; based on the range-dimensional FFT, static clutter removal, Capon azimuth angle spectrum estimation, target detection, spectrum expansion, parameter estimation, Capon pitch angle estimation, and Doppler estimation processing are performed to obtain point cloud information of the observation scene including distance, azimuth, pitch angle, Doppler, and signal-to-noise ratio.

[0044] Among them, a first-order recursive iterative calculation method is used to perform complex inter-frame clutter suppression on each range unit of the current frame one-dimensional range image.

[0045] The present invention provides an electronic device, which includes a processor and a memory for storing executable instructions of the processor; the processor is used to read the executable instructions from the memory and execute the instructions to implement the bed fall detection method for home health monitoring radar described in the present invention.

[0046] The present invention provides a computer-readable storage medium, which stores a computer program. The computer program is used to execute the bed-fall detection method for home health monitoring radar according to the present invention.

[0047] Beneficial effects

[0048] 1. The method presented in this paper is an innovative bed-fall detection technology for home health monitoring radars. This technology deeply integrates one-dimensional range profile (HRRP) features (such as RCS amplitude) with point cloud information (such as the number of clustered targets). By designing efficient feature mining methods and intelligent decision logic, it aims to effectively overcome the aforementioned challenges and achieve high-accuracy and low-false-alarm bed-fall detection in complex home environments. Specifically, it simultaneously utilizes two features: the number of point cloud clusters (ClusterNum) and the time series variation of radar cross-section (RCS) amplitude to determine bed-fall. The number of point clouds reflects the spatial dispersion of targets (point clouds disperse and their number may decrease after a fall), while the RCS amplitude reflects the target's reflection strength (a fall may cause the RCS to drop sharply due to distance from the radar or changes in posture). This multi-dimensional information fusion significantly improves the accuracy and robustness of bed-fall detection compared to single-feature detection, reducing false positives caused by environmental interference or local target motion.

[0049] 2. The present method incorporates a unique dynamic cumulative counting mechanism (AccFlag1 and AccFlag2) for RCS amplitude. This mechanism not only records the number of consecutive frames in which the RCS is below or above the threshold (Th1), but also counts down when the state reverses (e.g., from high RCS to low RCS or vice versa). This design effectively captures the continuous change of the target state (e.g., "in bed" to "out of bed"), avoids false triggering caused by transient fluctuations, and makes the detection logic more suitable for the persistent low RCS state characteristic of falls from bed.

[0050] 3. The detection method of this invention uses the number of clusters (ClusterNum) after point cloud aggregation as a key criterion. When a suspected fall from bed occurs (based on RCS dynamic counting), ClusterNum is further checked. If ClusterNum is below a threshold (Th3), indicating a small number of valid target points in the scene (e.g., a person scattered on the ground after falling, resulting in a single cluster of point clouds), the event is considered a fall from bed. Otherwise (ClusterNum is higher), the event is considered normal bedside or indoor activity (e.g., a person walking around the bed after getting out of bed may form multiple point cloud clusters). This feature significantly reduces the probability of misidentifying normal waking activity as a fall from bed (false alarms).

[0051] 4. The method of the present invention uses a first-order recursive clutter map method to suppress static and slowly varying clutter in the one-dimensional range image, effectively extracting the RCS amplitude of the real moving target, ensuring the accuracy and reliability of the RCS time series characteristics. A point cloud aggregation algorithm based on multi-dimensional (range, azimuth, pitch, and velocity) thresholds is designed to more accurately reflect the number of independent targets in the scene (ClusterNum), providing a reliable basis for distinguishing between falling out of bed and normal activities.

[0052] 5. The device used to implement the method of the present invention utilizes millimeter-wave radar technology, achieving completely contactless monitoring. This eliminates the need for the person being cared for to wear any equipment, making it convenient and comfortable to use. Furthermore, millimeter-wave radar does not capture optical image information, effectively protecting personal privacy and making it particularly suitable for use in home environments.

[0053] 6. The medium of the present invention is used to implement the method of the present invention. By integrating the point cloud clustering feature (ClusterNum) and the dynamic RCS change features (AccFlag1 / AccFlag2), combined with an innovative point cloud clustering algorithm and RCS dynamic counting mechanism, it achieves high-accuracy, low-false-alarm detection of bed falls in home environments. Its non-contact, privacy-preserving, real-time, and highly adaptable features make it of great application value and promising prospects in the field of home safety monitoring for specific populations such as the elderly and patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 Schematic diagram of the process of the present invention.

[0055] Figure 2 This is a time variation curve diagram of the number of point clouds during normal bed rest and falling-out events in the method of the present invention.

[0056] Figure 3 This is a diagram of the RCS changes and bed-fall detection results during normal bed rest and a bed-fall event in the method of the present invention.

[0057] Figure 4 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0059] This paper proposes a bed-fall detection method for a home health monitoring radar. The radar can be mounted on the wall 1 meter above the headboard of the bed in the monitored room, with the antenna beam center illuminating the center of the bed at a downward tilt angle. A linear frequency-modulated continuous wave (chirp) signal is transmitted in TDM-MIMO mode. The intermediate frequency echo data collected by the radar is rearranged through antenna channels, and the data from each receiving antenna is segmented to generate corresponding virtual antenna data. A three-dimensional radar data cube (radar cube) is generated, which contains the fast-time sampling points, the number of slow-time chirps, and the virtual antenna channel number.

[0060] The process of the present invention is as follows Figure 1 As shown. It includes the following steps:

[0061] Step 1: Perform one-dimensional range imaging and point cloud imaging on the raw echo data collected by the radar to obtain a one-dimensional range image (HRRP) and point cloud information (Cloud) of the observation scene;

[0062] Furthermore, in this embodiment, the echo sampling point of the first Chirp signal in each frame is taken to perform a fast Fourier transform (ie, range-dimensional FFT) to obtain a high-resolution one-dimensional range profile (HRRP) of the observed scene.

[0063] Based on the range-dimensional FFT, further basic processing such as static clutter removal, Capon azimuth angle spectrum estimation, target detection (CFAR), spectrum expansion, parameter estimation, Capon pitch angle estimation, and Doppler estimation is performed to obtain point cloud information (Cloud) of the observation scene including distance, azimuth, pitch angle, Doppler, and signal-to-noise ratio.

[0064] Step 2: Use the clutter map method to suppress clutter on the acquired one-dimensional range profile (HRRP). Calculate the amplitude of the cleaned range spectrum data and perform a peak search to find the point with the maximum amplitude. This point's value is used as the target's amplitude, and the RCS amplitude time series (RCS) of the observed scene is obtained. Specifically, a peak search is performed on the clutter-suppressed one-dimensional range profile data. The range bin with the maximum amplitude (RCS) and the corresponding amplitude are used as the target's range bin (Rbin) and the target's amplitude (RCS).

[0065] The first-order recursive iterative calculation method is used to perform complex inter-frame clutter suppression on each range unit of the current frame one-dimensional range image (HRRP). The specific method of clutter suppression is as follows:

[0066] 1) Background clutter estimation:

[0067] (1)

[0068] In the formula is the iteration factor of clutter estimation, is the observed value of the real part of the current distance unit, is the observation value of the imaginary part of the current distance unit, is the historical estimate of the real part of the clutter in this range cell; is the historical estimate of the imaginary part of the clutter in this range cell. is the estimated value of the real part of the clutter in the range unit; is the estimated value of the imaginary part of the clutter in the range unit;

[0069] 2) Clutter cancellation:

[0070] (2)

[0071] In the formula RCS is the amplitude of the range unit after clutter suppression;

[0072] 3) Update the clutter history value:

[0073] (3)

[0074] Step 3: Perform feature mining on the amplitude time series RCS and count the number of continuous frames AccFlag1 below the threshold Th1 (typical value is 100) and the number of continuous frames AccFlag2 above the threshold Th1 (typical value is 100);

[0075] Step 4: Agglomerate the acquired point cloud information and obtain the number of clusters after the point cloud agglomeration, ClusterNum; specifically, agglomerate the point cloud of the current frame and count the number of clusters after the agglomeration, ClusterNum (the initial value is 0). The specific method of point cloud agglomeration is as follows:

[0076] Step 1: Assume that the point cloud set is obtained There are K points in total. Take the first point as the reference point. Assume that the coordinates of this point are ,in: is the signal-to-noise ratio, For distance, is the azimuth, is the pitch angle, is the Doppler velocity. And the label of each point is initialized as: GroupNo = [1, zeros(1,K - 1)], where the first point is the reference point and its label is 1, and the labels of the remaining K - 1 points are 0;

[0077] Step 2: Calculate the reference points according to the following formula (4) Divide by point cloud collection Other points The Euclidean distance.

[0078] (4)

[0079] Where: Represents the absolute value. is a vector containing five-dimensional information: distance, azimuth, pitch angle, and Doppler velocity, where Represents the point cloud distance difference; Represents the point cloud azimuth difference; Represents the pitch angle difference of the point cloud; Represents the Doppler velocity difference of the point cloud. and The five-dimensional information difference of distance, azimuth, pitch angle and Doppler speed is used to make a threshold decision. If the decision condition is met, Points associated with reference points , set the label of the point to 1.

[0080] The judgment conditions are as follows: the point cloud distance difference is less than the point cloud distance difference threshold (typical value is 0.1), and the point cloud azimuth difference is less than the point cloud azimuth difference threshold (typical value is 15), and the point cloud pitch angle difference is less than the point cloud pitch angle difference threshold (typical value is 15), and the Doppler velocity difference is less than the point cloud Doppler velocity difference threshold (typical value is 2).

[0081] Step 3: Find all points with label 1 in the label set GroupNo and count the number of these points N. If the number of points N is greater than the minimum point cloud number threshold TargetNum (typical value is 8), the point set with label 1 is considered to be a valid target point set, and the number of classes ClusterNum is increased by 1.

[0082] Step 4: Collect in point cloud In the , remove the points that have been associated and assigned to the same class, and extract the set of remaining unassigned points. That is, find all the points with a label of 0 in GroupNo, and use these points as the new point set pair Update and repeat the above steps until all points are condensed and classified, and obtain the number of condensed classes ClusterNum.

[0083] Step 5: Determine the abnormal falling-out-of-bed behavior based on the RCS features and the number of clusters after point cloud aggregation (ClusterNum) to determine whether a falling-out-of-bed event has occurred.

[0084] Based on the modeling of falling-out-of-bed behavior and the analysis of false alarms, the “fall-out-of-bed” considered in this invention is divided into the following three situations:

[0085] (1) If a person is in bed and then accidentally falls off the bed, this situation is considered an abnormal situation and needs to be detected promptly and accurately.

[0086] (2) A person faints or slips while sitting on the edge of the bed or standing up from the bed; this situation is also considered an abnormal situation and needs to be detected promptly and accurately.

[0087] (3) A person gets up normally and then moves around the bed or around the house; this situation is considered normal and the false alarms of such situations need to be reduced.

[0088] Therefore, based on the modeling of falling-out-of-bed behavior, the embodiments of the present invention mainly implement the determination of abnormal falling-out-of-bed behavior and the suppression of false alarms by designing logical judgment conditions based on the RCS change characteristics and the target height change characteristics. The specific implementation method is as follows:

[0089] (1) Dynamically count frames with RCS values ​​below and above the threshold Th1 (typical value is 100), where AccFlag1 represents the number of frames with RCS values ​​below Th1; AccFlag2 represents the number of frames with RCS values ​​above Th1. The specific implementation of dynamic counting is as follows:

[0090] 1) If the RCS value of the current frame is less than the threshold Th1, AccFlag1 is incremented by 1 and AccFlag2 is checked. If it is greater than 0, AccFlag2 is decremented by 1; otherwise, AccFlag2 is cleared to 0.

[0091] 2) If the current frame RCS value is greater than the threshold Th1, AccFlag2 is incremented by 1, and AccFlag1 is checked. If it is greater than 0, AccFlag1 is decremented by 1, otherwise AccFlag1 is cleared to 0.

[0092] (2) Design logical judgment conditions to perform fall detection and false alarm suppression. If AccFlag1 > Th2 and AccFlag2 > Th2, it is considered that there is a suspected fall behavior. On this basis, further check the number of classes ClusterNum after the point cloud condensation of the current frame. If ClusterNum < Th3, it is determined that a fall behavior has occurred, and the fall flag is output; otherwise, it is considered normal activity beside the bed or inside the house. The typical value of Th2 is 80; the typical value of Th3 is 2.

[0093] To illustrate the effectiveness of the present invention, the following experimental verification is carried out. The experimental data uses the measured radar echo data, and the experimental operation results are as Figure 2 and Figure 3 shown. Figure 2 It is a graph of the change in the number of point clouds over time when a fall event occurs; Figure 3 It is a curve graph of the RCS change and the fall detection result after the fall event occurs. This experiment verifies the effectiveness of the fall detection technology through the method proposed in the present invention. From the experimental results, it can be seen that the fall detection technology provided by the present invention can be applied to home environments with obvious differences, accurately detect the occurrence of fall events, and then alarm for abnormal out-of-bed behaviors.

[0094] This embodiment of the present application also provides an electronic device, Figure 4The structure of an electronic device provided in an embodiment of the present invention is shown. For example, the electronic device 40 may include a processor 41, a memory 42, and a transmission device 43. The processor is used to execute the bed-fall detection method for home health monitoring radar mentioned in the above embodiment, wherein the processor and memory may be connected via a bus or other means, with bus connection being used as an example. The transmission device may be connected to the processor and memory via a wired or wireless connection. The memory, as a non-transitory computer-readable storage medium, may be used to store non-transitory software programs, non-transitory computer executable programs, and modules, such as the program instructions / modules corresponding to the bed-fall detection method for home health monitoring radar in the embodiment of the present application. The processor executes the various functional applications and data processing of the processor by running the non-transitory software programs, instructions, and modules stored in the memory, thereby implementing the bed-fall detection method for home health monitoring radar in the above method embodiment. The memory may include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for the function; the data storage area may store data created by the processor, etc. In addition, the memory may include a high-speed random access memory and may also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include a memory remotely located relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the aforementioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof. The one or more modules are stored in the memory, and when executed by the processor, the method for detecting falling out of bed for a home health monitoring radar in the embodiment is executed.

[0095] As another aspect, the present application also provides a computer-readable storage medium, which may be the computer-readable storage medium included in the device described in the above embodiment; or it may be a computer-readable storage medium that exists independently and is not assembled into the device. The computer-readable storage medium may be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the art. The computer-readable storage medium stores one or more programs, which are used by one or more processors to execute the bed-fall detection method for home health monitoring radar described in the present application.

[0096] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for detecting falling out of bed for home health monitoring radar, characterized in that: include; The original echo data collected by the radar is processed by one-dimensional range imaging and point cloud imaging to obtain a one-dimensional range image and point cloud information of the observation scene; clutter suppression is performed on the one-dimensional range image, the amplitude of the distance spectrum data after de-cluttering is calculated and a peak search is performed to find the point with the maximum amplitude. The value of this point is used as the amplitude of the target, and the RCS amplitude time series of the observation scene is obtained. Feature mining is performed on the amplitude time series RCS, and the number of continuous frames AccFlag1 below the threshold Th1 and the number of continuous frames AccFlag2 above the threshold Th1 are counted; the obtained point cloud information is condensed to obtain the number of clusters after the point cloud condensation (ClusterNum); based on the RCS features and the number of clusters after the point cloud condensation (ClusterNum), abnormal falling behavior is judged to determine whether a falling event has occurred.

2. The method according to claim 1, wherein The specific implementation method for judging abnormal falling-out-of-bed behavior is as follows: Dynamically count frames with RCS values ​​lower than and higher than the threshold Th1, where AccFlag1 represents the number of frames with RCS values ​​lower than Th1; AccFlag2 represents the number of frames with RCS values ​​higher than Th1. The specific implementation of dynamic counting is: 1) If the RCS value of the current frame is less than the threshold Th1, AccFlag1 is incremented by 1, and AccFlag2 is checked. If it is greater than 0, AccFlag2 is decremented by 1, otherwise AccFlag2 is cleared to 0; 2) If the RCS value of the current frame is greater than the threshold Th1, AccFlag2 is incremented by 1, and AccFlag1 is checked. If it is greater than 0, AccFlag1 is decremented by 1, otherwise AccFlag1 is cleared to 0; Design logical judgment conditions to detect falling out of bed and suppress false alarms. If AccFlag1>Th2 and AccFlag2>Th2, it is considered that there is a suspected falling out of bed behavior. On this basis, check the number of clusters after the point cloud of the current frame (ClusterNum). If ClusterNum is less than the threshold Th3, it is determined that the falling out of bed behavior has occurred and the falling out of bed flag is output; otherwise, it is considered that the person is moving normally beside the bed or in the room.

3. The method according to claim 2, wherein ,The specific method of point cloud condensation is as follows: Step 1: Assume that the obtained point cloud set z total There are K points in total. Take the first point as the reference point. Assume that the coordinates of this point are z0 = (snr, r, a, e, v), where: snr is the signal-to-noise ratio, r is the distance, a is the azimuth, e is the pitch angle, and v is the Doppler velocity. i , i=1,2…K-1; initialize the label of each point to: GroupNo=[1,zeros(1,K-1)], where the first point is the reference point, its label is 1, and the initial labels of the remaining K-1 points are 0; Step 2: Calculate the reference point z0 and the rest of the points z in the point cloud set except z0 respectively i The Euclidean distance is as follows: With diff =abs(z0-z i ) =(|snr0-snr i |,|r0-r i |,|a0-a i |,|e0-e i |,|v0-v i |) In the formula: || represents the absolute value, z diff is a vector containing the five-dimensional information of signal-to-noise ratio, distance, azimuth, pitch angle, and Doppler velocity, where |snr0-snr i | represents the difference in signal-to-noise ratio of the point cloud; |r0-r i | represents the point cloud distance difference; |a0-a i | represents the point cloud azimuth difference; |e0-e i | represents the pitch angle difference of the point cloud; |v0-v i | represents the Doppler velocity difference of the point cloud; z0 and z i The distance, azimuth, pitch angle, and Doppler velocity difference between them are used for threshold judgment. If the judgment conditions are met, z i point is associated with the reference point z0, and z i The label of the point is set to 1, and the label set GroupNo is updated; The judgment condition is as follows: the point cloud distance difference is less than the point cloud distance difference threshold, the point cloud azimuth difference is less than the point cloud azimuth difference threshold, the point cloud pitch angle difference is less than the point cloud pitch angle difference threshold, and the point cloud Doppler velocity difference is less than the point cloud Doppler velocity difference threshold; Step 3: Find all points with label 1 in the updated label set GroupNo and count the number of these points N; if the number of points N is greater than the minimum point cloud number threshold TargetNum, the point set with label 1 is considered to be a valid target point set, and the number of classes ClusterNum is increased by 1; Step 4: In the point cloud collection z total In the process, remove the points that have been associated with the same class, extract the set of remaining unassigned points, repeat the above steps until all points are condensed and classified, and obtain the number of condensed classes ClusterNum.

4. The method according to any one of claims 1 to 3, wherein ,The clutter map method is used to suppress clutter on the one-dimensional range image of the ,observation scene.

5. The method according to any one of claims 1 to 3, wherein: Take the echo sampling point of the first Chirp signal in each frame and perform fast Fourier transform to obtain the one-dimensional range image of the observation scene; Based on the range-dimensional FFT, static clutter removal, Capon azimuth angle spectrum estimation, target detection, spectrum expansion, parameter estimation, Capon pitch angle estimation, and Doppler estimation are performed to obtain point cloud information of the observation scene including distance, azimuth, pitch angle, Doppler, and signal-to-noise ratio.

6. The method according to claim 4, wherein The first-order recursive iterative calculation method is used to suppress the complex inter-frame clutter of each range unit of the current frame one-dimensional range image.

7. An electronic device, characterized in that: The electronic device includes a processor and a memory for storing executable instructions of the processor; the processor is used to read the executable instructions from the memory and execute the instructions to implement the bed fall detection method for home health monitoring radar as described in any one of claims 1-6 above.

8. A computer-readable storage medium, characterized in that The storage medium stores a computer program, which is used to execute the bed-fall detection method for home health monitoring radar described in any one of claims 1 to 6.

Images