CN-121978679-A - Fall detection device and method for low-height millimeter wave radar

CN121978679ACN 121978679 ACN121978679 ACN 121978679ACN-121978679-A

Abstract

The invention relates to the technical field of radar detection and discloses a low-height millimeter wave radar falling detection device which comprises a multi-antenna array module, a narrow-beam low-elevation radar unit, an anti-interference component and a signal processing module. The low-altitude millimeter wave radar fall detection method comprises a dynamic background modeling unit, a micro Doppler characteristic extraction unit and a micro Doppler frequency shift characteristic extraction unit, wherein the dynamic background modeling unit is used for constructing a Gaussian mixture model through static scanning, carrying out pixel-level background difference in real time and judging a foreground target, and the micro Doppler characteristic extraction unit is used for carrying out short-time Fourier transform on echo signals of a human body, generating a time-frequency diagram and extracting micro Doppler frequency shift characteristics. Compared with the prior art, the invention solves the problem of low detection precision caused by environmental interference when the conventional millimeter wave radar is installed at a low height, realizes accurate identification of falling of a human body at the installation height of less than or equal to 80cm, and is compatible with complex indoor environments.

Inventors

CHEN XIAOWEI
ZHANG DALIN
ZHANG YAO

Assignees

碳基脉冲(深圳)科技有限公司

Dates

Publication Date: 20260505
Application Date: 20250820

Claims (7)

1. A low-height millimeter wave radar fall detection device, comprising: The multi-antenna array module adopts a 4Tx8Rx MIMO architecture, the antenna spacing is lambda/2, the multi-antenna array module is integrated in a rectangular metal cavity (3) with the height of 3cm, and a focused beam pointing to the height of the waist of a human body is dynamically generated through a beam forming algorithm; the narrow beam low elevation radar unit has a central frequency of 77GHz, a beam angle of 12 degrees, and a physical installation angle of 15 degrees inclined upwards, so that the central axis of the beam is aligned with the waist of a human body, and the detection range is limited in a cylindrical area with a distance radar of 0.3-3 meters and a vertical height of 0.5-1.8 meters; The anti-interference assembly comprises a metal shielding cover (1) and a radio frequency filter circuit (2), wherein the metal shielding cover (1) wraps the bottom surface and two side surfaces of the radar module, only a forward 120-degree detection window is reserved, and the radio frequency filter circuit (2) is integrated with a 50-90GHz band-pass filter; and the signal processing module is used for executing dynamic background modeling, micro Doppler feature extraction, three-dimensional gesture calculation and falling judgment.
2. A low-height millimeter wave radar fall detection method according to claim 1, wherein the signal processing module comprises: the dynamic background modeling unit builds a Gaussian mixture model through static scanning, carries out pixel-level background difference in real time, and judges a foreground target; the micro Doppler characteristic extraction unit is used for carrying out short-time Fourier transform on the human echo signals, generating a time-frequency diagram and extracting micro Doppler frequency shift characteristics; The three-dimensional attitude resolving unit is used for generating a three-dimensional point cloud based on virtual aperture synthesis and a back projection algorithm of the multi-antenna array, extracting a joint point through self-adaptive radius filtering and human body structure priori, and calculating a pitch angle, a roll angle and a joint angular velocity; And the falling judgment unit adopts an SVM classifier, and integrates the pitch angle, the joint angular velocity and the micro Doppler entropy value characteristics to carry out falling judgment.
3. The low-altitude millimeter wave radar fall detection method according to claim 2, wherein the dynamic background modeling unit specifically comprises: The initialization stage comprises the steps of collecting radar echoes of static objects such as ground, furniture and the like through static scanning for 10 minutes, constructing a Gaussian mixture model, and enabling each background pixel to correspond to 3 Gaussian components; And (3) carrying out pixel-level background difference on the current frame signal, calculating D (x) = |f (x) -B (x) |/sigma (x), and judging that the current frame signal is a foreground target (human body signal) if D (x) >2.5, or else, judging that the current frame signal is background clutter.
4. The low-altitude millimeter wave radar fall detection method according to claim 2, wherein the micro-doppler feature extraction unit specifically comprises: STFT is carried out on the human echo signals, and a time-frequency diagram is generated; And identifying micro Doppler frequency shift corresponding to the swing of the limbs by adopting a peak detection algorithm, and constructing a feature vector V= [ f1, f2, ], wherein fi is the ith micro motion component frequency.
5. The low-altitude millimeter wave radar fall detection method according to claim 2, wherein the three-dimensional attitude calculation unit specifically comprises: first, three-dimensional point cloud reconstruction and denoising The virtual aperture synthesis based on the 4Tx8Rx MIMO array comprises the steps of transmitting linear frequency modulation continuous wave signals through a transmitting antenna in a time sharing way, synchronously collecting echoes through a receiving antenna, calculating a target azimuth angle and a pitch angle by utilizing the phase difference of each antenna pair, synthesizing a 32X 32 virtual array, and improving the spatial resolution to 10cm X10 cm; Imaging by a back projection algorithm, namely focusing echoes of all channels to a three-dimensional space grid through a BP algorithm after carrying out distance-Doppler processing on virtual array data to generate an initial point cloud; the formula is P (x, y, z) = Σi=132 Σj=132 si, j (r, Φ) ∈ (-jλ4pi (xsin θaz, i+ ycos θaz, i+ zsin θel, j)); wherein Si, j (r, phi) is the distance-Doppler spectrum of the ith row and jth column virtual antennas, and r is 77GHz millimeter wave wavelength; dynamic noise filtering, namely adopting adaptive radius filtering aiming at clutter point clouds formed by ground reflection in a low-altitude scene: 1) Setting a radius threshold R=20cm, and counting the number of neighbors in 20cm around each point cloud; 2) If the number of neighbors is less than 5, judging clutter points and eliminating, and reserving effective human body point clouds; Secondly, extracting human body characteristic points; Based on the region division of human body structure prior, the point cloud is divided into 3 regions according to the height by utilizing the characteristic that human body targets are mainly distributed in z=0.5-1.8m in a low-height scene: 1) The upper body region (z=1.2-1.8 m) contains characteristic points such as shoulder joint and neck; 2) Torso region (z=0.8-1.2 m) containing characteristic points such as hip joint and waist; 3) The lower limb region (z=0.5-0.8 m) comprises characteristic points such as knee joint, ankle joint and the like; Clustering and screening the characteristic points: Performing European clustering on the point cloud of each region, and extracting a clustering center as a candidate feature point; screening out candidate points conforming to human body structure constraint by combining with skeleton dynamics model, and determining final joint point coordinates (such as Knee joint Shoulder joint ; Third, attitude parameter calculation Calculating a pitch angle, namely calculating an included angle between the human body longitudinal axis and the vertical direction (z axis) by taking the human body longitudinal axis as a reference through a vector dot product: ; aiming at the problem that human body forward inclination is easy to be blocked in a low-height scene, when a shoulder joint point is missing, a hip joint and knee joint connecting line is automatically started to replace a longitudinal axis for calculation; Roll angle calculation based on projection of hip joint H onto horizontal plane Projection with shoulder joint S Calculating the included angle between the connecting line and the x axis: ; joint angular velocity calculation, namely differentiating joint point coordinates of 3 continuous frames, and calculating the angular velocity through a center difference method: ; Wherein, the , As the joint angle of the adjacent frames, =0.1 S, ensuring a capture accuracy of ±5°/s for rapid posture change at the time of fall.
6. The low-altitude millimeter wave radar fall detection method according to claim 2, wherein the fall discrimination unit specifically comprises: Setting a falling characteristic threshold value: Pitch angle θ >45 ° and duration >0.3s; joint angular velocity ω >150 °/s (knee or hip); the micro Doppler entropy value E is more than 3.5; adopting an SVM classifier, fusing multiple feature vectors to make decisions, and judging functions as follows: f (x) =w1θ+w2ω+w3e+b, where w1=0.4, w2=0.35, w3=0.25, b= -2.8.
7. The low-altitude millimeter wave radar fall detection method according to claim 2, further comprising an adaptive calibration mechanism, comprising: initial calibration, namely automatically executing environment scanning after installation, recording furniture positions and generating an ROI mask; On-line update-triggering background model update once per hour, adapting to environmental changes (e.g. furniture movement, floor carpeting resulting in reflectance changes) through incremental GMM algorithm (μ_new=μ_old+α (x- μ_old)).

Description

Fall detection device and method for low-height millimeter wave radar Technical Field The invention relates to the technical field of radar detection, in particular to a low-height millimeter wave radar falling detection device and method. Background Millimeter wave radars are widely applied to the fields of intelligent security and health monitoring due to the characteristics of strong penetrability, excellent anti-interference capability, no influence of illumination and the like. Traditional fall detection schemes rely mainly on visual sensors or millimeter wave radars mounted at high altitudes (1.8 meters or more). The vision scheme has privacy leakage risk, and although the traditional millimeter wave radar can acquire human body posture data through overlooking angles when installed at a high place, the following problems exist in a low-altitude (< 80 cm) installation scene: ground reflection causes strong clutter, and human echo signals are submerged; furniture shielding forms a monitoring blind area; it is difficult to distinguish between low-lying movements (e.g. bending over) and falling postures. In the prior art, millimeter wave radars installed at low heights cannot meet the requirements of small-sized houses, nursing bedside and other scenes due to low signal-to-noise ratio, insufficient detection precision and high false detection rate. Accordingly, there is a need for an apparatus and method that enables fall detection with high accuracy in a low-height installation environment. Disclosure of Invention The invention aims to solve the technical problems of overcoming the technical difficulties, and providing the low-height millimeter wave radar falling detection device and method, which solve the problem of low detection precision caused by environmental interference when the conventional millimeter wave radar is installed at a low height, realize accurate identification of falling of a human body at the installation height of less than or equal to 80cm, and are compatible with complex indoor environments (such as dense furniture and various ground materials). In order to solve the technical problems, the technical scheme provided by the invention is as follows: a low-height millimeter wave radar fall detection device, comprising: the multi-antenna array module adopts a 4Tx8Rx MIMO architecture, the antenna spacing is lambda/2 (lambda is millimeter wave wavelength), the antenna is integrated in a rectangular metal cavity with the height of 3cm, and a focused beam pointing to the height of the waist of a human body is dynamically generated through a beam forming algorithm; the narrow beam low elevation radar unit has a central frequency of 77GHz, a beam angle of 12 degrees, and a physical installation angle of 15 degrees inclined upwards, so that the central axis of the beam is aligned with the waist of a human body, and the detection range is limited in a cylindrical area with a distance radar of 0.3-3 meters and a vertical height of 0.5-1.8 meters; the anti-interference assembly comprises a metal shielding cover and a radio frequency filter circuit, wherein the metal shielding cover wraps the bottom surface and two side surfaces of the radar module, only a forward 120-degree detection window is reserved, and the radio frequency filter circuit 2 is integrated with a 50-90GHz band-pass filter; and the signal processing module is used for executing dynamic background modeling, micro Doppler feature extraction, three-dimensional gesture calculation and falling judgment. The low-height millimeter wave radar fall detection method, the signal processing module comprises: the dynamic background modeling unit constructs a Gaussian Mixture Model (GMM) through static scanning, carries out pixel-level background difference in real time, and judges a foreground target; The micro Doppler characteristic extraction unit is used for carrying out short-time Fourier transform (STFT) on the human echo signals, generating a time-frequency diagram and extracting micro Doppler frequency shift characteristics; The three-dimensional attitude resolving unit is used for generating a three-dimensional point cloud based on virtual aperture synthesis and a back projection algorithm of the multi-antenna array, extracting a joint point through self-adaptive radius filtering and human body structure priori, and calculating a pitch angle, a roll angle and a joint angular velocity; And the falling judgment unit adopts an SVM classifier, and integrates the pitch angle, the joint angular velocity and the micro Doppler entropy value characteristics to carry out falling judgment. Compared with the prior art, the invention has the advantages that: 1. The invention improves the clutter suppression ratio by 25dB through multi-antenna beam forming and the metal shielding case 1; 2. The narrow beam low elevation design of the invention leads the ground reflection signal attenuation to be more than or equal to 40%; 3. The three-dimensi