60G millimeter wave radar personnel falling detection system and method applied to intelligent closestool

Abstract

The invention relates to the technical field of smart home and health monitoring, in particular to a 60G millimeter wave radar personnel tumble detection system and method applied to an intelligent closestool, and the system comprises a detection module, a signal processing module, a coordinate conversion module, a grid processing module, a neural network processing module, a tumble judgment module and a wireless communication module. The detection module is installed at the bottom of a closestool in a hole-opening embedding mode and comprises two 60G millimeter wave radar modules and an intelligent closestool ring pressure sensor module. A three-dimensional point cloud is generated by acquiring radar echo signals, point cloud data is gridded and then input into a full-connection neural network for processing, tumble state judgment is carried out in combination with pressure sensor data, finally, a result is transmitted to a mobile phone APP in real time through a Bluetooth module, high-precision non-contact tumble detection is carried out, and the tumble detection precision is improved. The method has the advantages of privacy protection, high environmental adaptability, no need of wearing equipment and the like, and solves the problems of privacy leakage, high false alarm rate, poor environmental adaptability and the like in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical fields of smart home and health monitoring, and particularly to a personnel fall detection technology based on millimeter-wave radar. More specifically, it is a 60G millimeter-wave radar personnel fall detection system and method applied to a smart toilet. This method and system combine the high-precision non-contact detection ability of millimeter-wave radar with the pressure sensor of the smart toilet to achieve real-time monitoring of the personnel's fall state. Background Art

[0002] Falling is a major problem in the health monitoring of the elderly. Especially in environments such as home bathrooms, due to slippery floors, narrow spaces, and inconvenient movements, the elderly are more prone to falling accidents. Falling incidents may cause serious physical injuries and even endanger lives. Therefore, developing a technology that can detect the falling condition in real time and give an alarm in time is of great significance for reducing accidental injuries and saving lives.

[0003] Existing fall detection technologies are mainly divided into the following categories: First, there are methods based on wearable devices, such as smart bracelets and acceleration sensors. These devices judge whether a fall has occurred by monitoring the human body's motion state. However, they rely on the user to actively wear them. If the user forgets or refuses to wear them, the detection cannot be carried out, and the risk of device damage is relatively high. Second, there are monitoring methods based on cameras. This method judges whether a person has fallen through image recognition technology. However, this method has serious problems of privacy leakage, especially in sensitive areas such as bathrooms, and the detection accuracy of cameras is relatively low in low-light conditions. In addition, there are also some detection methods based on infrared or ultrasonic waves. Although such technologies can achieve a certain degree of non-contact detection, due to low resolution and poor environmental adaptability, they are often restricted by the usage scenarios.

[0004] Millimeter-wave radar, as an emerging non-contact sensing technology, has received extensive attention in recent years, especially in the field of health monitoring, where it has significant advantages. The millimeter-wave radar in the 60GHz band has extremely high spatial resolution and can accurately capture the position and posture changes of personnel. At the same time, the millimeter-wave radar works based on the reflection characteristics of electromagnetic waves and does not record video or image data, so it can effectively protect user privacy. In addition, the millimeter-wave radar is not sensitive to environmental factors such as light and humidity and is very suitable for personnel detection in complex environments. Compared with traditional technologies, millimeter-wave radar has obvious advantages in privacy protection, detection accuracy, and environmental adaptability.

[0005] As a smart home device with an increasing penetration rate, intelligent toilets have integrated various sensor technologies, such as pressure sensors and temperature sensors. These sensors can monitor the user's usage status in real time and provide additional auxiliary information for fall detection. For example, the pressure sensor can determine whether the user touches the toilet seat, which provides strong support for identifying the person's status. However, relying solely on the sensors of the intelligent toilet cannot comprehensively cover fall events. Therefore, combining millimeter-wave radar with the sensor information of the intelligent toilet can make up for the deficiencies of single technology and achieve higher detection accuracy. Summary of the Invention

[0006] The purpose of the present invention is to provide a 60G millimeter-wave radar personnel fall detection system and method applied to intelligent toilets, aiming to solve problems such as privacy leakage, high false alarm rate, and poor environmental adaptability existing in existing fall detection technologies. By combining millimeter-wave radar with the pressure sensor of the intelligent toilet, a neural network model is used to detect and judge the personnel fall status in real time, and finally the result is displayed through a mobile APP terminal, providing timely and effective health monitoring services for users.

[0007] The present invention proposes a 60G millimeter-wave radar personnel fall detection system applied to intelligent toilets, including:

[0008] A detection module, which is installed at the bottom of the toilet in an opening-embedded manner, including two 60G millimeter-wave radar modules and an intelligent toilet seat pressure sensor module;

[0009] A signal processing module, connected to the detection module, for processing the echo data of multiple chirp signals collected by the two 60G millimeter-wave radar modules. The processing includes performing discrete Fourier transform in the fast time dimension and the slow time dimension, and using digital beamforming technology to obtain the horizontal angle and elevation angle of the target, and generating three-dimensional point cloud information of the target;

[0010] A coordinate conversion module, connected to the signal processing module, for converting the point cloud target in the radar coordinate system to the world coordinate system and mapping these point clouds to a specific area in three-dimensional space;

[0011] A grid processing module, connected to the coordinate conversion module, for performing grid processing on the three-dimensional space data, dividing the space into multiple cells, taking the number of point clouds as the value of the space unit, and arranging these values as a column vector according to the spatial position;

[0012] A neural network processing module, connected to the grid processing module, for using the spatial vector value as the input of a fully connected neural network to process and train the grid, obtaining network weights, and deploying the trained network to a processor;

[0013] A fall determination module, connected to the neural network processing module and the intelligent toilet seat pressure sensor module, is configured to determine the fall status of a person by combining the fall probability value output by the fully connected neural network and the data of the intelligent toilet seat pressure sensor;

[0014] A wireless communication module, connected to the fall determination module, is configured to send the determination result of the fall detection system to a mobile APP terminal for result display.

[0015] Preferably: The detection module is installed outside the intelligent toilet by means of embedding through an opening.

[0016] Preferably: The coordinate conversion module performs the conversion from the radar coordinate system to the world coordinate system according to the following formula:

[0017]

[0018] Where (r, θ, Φ) represents the polar coordinates of the target point cloud in the radar coordinate system, where r represents the distance, θ represents the horizontal angle, and Φ represents the pitch angle;

[0019] (x w , y w , z w ) represents the coordinate values of the target point cloud in the world coordinate system;

[0020] (γ, β, γ) represents the rotation angles of the radar around the x, y, and z axes;

[0021] x offset y offset , z offset represents the displacement amounts of the radar on the x, y, and z axes.

[0022] Preferably: The grid processing module operates according to the following steps:

[0023] Based on the 3D point cloud data collected by the millimeter-wave radar, find the minimum and maximum coordinate values of all points to determine the ranges of the target area in the x, y, and z directions;

[0024] Evenly divide the determined spatial range into small cubic grids of the same size, and the size of each grid cell can be set according to the required resolution;

[0025] For each point in the point cloud data, find which grid cell it belongs to according to its x, y, and z coordinate values;

[0026] For each grid cell, count the number of points belonging to the grid, and at the same time count and calculate the signal-to-noise ratio, average reflection intensity, central position, etc. of these points to provide useful information for subsequent processing;

[0027] Grid data in the form of an output vector, where each grid contains its spatial position and statistical information.

[0028] Preferably, the fully connected neural network in the neural network processing module is trained according to the following formula:

[0029] f(G) = σ(W3 · ReLU(W2 · ReLU(W1 · G + b1) + b2)),

[0030]

[0031] where W1, W2, and W3 are the weight matrices of the fully connected layers, b1 and b2 are the bias terms, σ(·) represents the activation function, and the output value f(G) is the probability of falling. L is the cross-entropy loss value during network training.

[0032] Preferably, the fall determination module determines the fall status of a person according to the following steps:

[0033] Obtain the fall probability value P output by the fully connected network radar and the pressure sensor value P of the intelligent toilet seat pressure ;

[0034] Define the threshold T of the pressure value pressure , if P pressure > T pressure , it means that someone is touching the toilet seat;

[0035] Define the weight parameters w radar and w pressure , and calculate the fusion probability

[0036] If P combined ≥ T combined , it is determined to be in a fall state, and the system sends the result to the mobile APP for display through the wireless communication module.

[0037] Preferably, the processor is an ESP32 chip.

[0038] Preferably, the w radar is set to 0.7, the w pressure is set to 0.3, P combined > 0.8 indicates that the user has fallen, P combined ≤ 0.5 indicates that the target has not fallen, 0.8 ≤ P combined <0.5 indicates an undetermined state, and further determination needs to be combined with multi-frame data.

[0039] Preferably, the wireless communication module is a Bluetooth module.

[0040] A 60G millimeter-wave radar personnel fall detection method applied to a smart toilet, the method comprising the following steps:

[0041] Install two 60G millimeter-wave radar modules at the bottom of the toilet in an open-hole embedded installation manner;

[0042] Obtain echo data of multiple chirp signals through the two 60G millimeter-wave radar modules, perform discrete Fourier transform on them in the fast time dimension and slow time dimension, and use digital beam synthesis technology to obtain the horizontal angle and pitch angle of the target, and generate three-dimensional point cloud information of the target;

[0043] Convert the point cloud target in the radar coordinate system to the world coordinate system, and map these point clouds to a specific area in three-dimensional space;

[0044] Perform grid processing on the mapped three-dimensional space data, divide the space into multiple cells, and use the number of point clouds as the value of the space unit, and arrange them as column vectors according to the spatial position;

[0045] Use the spatial vector value as the input of a fully connected neural network for processing and training to obtain the weights of the network;

[0046] Deploy the trained network to an ESP32 processor, and generate a spatial vector in real time through the millimeter-wave radar as the input of the network to calculate the probability value of personnel falling;

[0047] Combine the fall probability value with the pressure sensor data of the smart toilet seat to determine the fall status of the personnel;

[0048] Send the determination result of the fall detection system to the mobile APP terminal through the Bluetooth module for result display.

[0049] The system and method of the present invention have the following beneficial effects:

[0050] 1. Strong privacy protection: The present invention adopts millimeter-wave radar technology, does not record video or image data, effectively protects user privacy, and is especially suitable for use in sensitive environments such as bathrooms.

[0051] 2. High detection accuracy: Through the high spatial resolution of the 60G millimeter-wave radar and the multi-sensor fusion technology of the smart toilet pressure sensor, the accuracy of fall detection is greatly improved, and the false alarm rate is effectively reduced.

[0052] 3. Good environmental adaptability: Millimeter-wave radar is not sensitive to environmental factors such as light and humidity, can work stably in complex environments, and can adapt to various lighting conditions and spatial layouts.

[0053] 4. No device to wear: The system adopts a non-contact detection method, and users do not need to wear any devices, which greatly improves the convenience and acceptance of use.

[0054] 5. Strong real-time performance: The system can process radar data in real time and perform fall determination to ensure that relevant personnel can be notified in time when a fall occurs.

[0055] 6. Convenient installation method: Adopting an embedded installation method with holes, it can be easily integrated into existing smart toilets without affecting the appearance and normal use of the toilets.

[0056] 7. Edge computing ability: Deploy the trained neural network model on the ESP32 chip to achieve real-time inference at the edge, reducing system latency and dependence on the cloud.

[0057] 8. Good scalability: The system architecture has good scalability, and parameters and models can be adjusted according to different scenario requirements, and it can also be extended to other home health monitoring scenarios for use. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a schematic diagram of the system architecture in an embodiment of the present invention;

[0059] Figure 2 It is a schematic diagram of the installation method and detection range of the detection system in an embodiment of the present invention;

[0060] Figure 3 It is a flowchart of the method in an embodiment of the present invention;

[0061] Figure 4a It is a schematic diagram of the spatial distribution of the 3D point cloud of a person generated by the millimeter-wave radar in an embodiment of the present invention Figure 1 ;

[0062] Figure 4b It is a schematic diagram of the spatial distribution of the 3D point cloud of a person generated by the millimeter-wave radar in an embodiment of the present invention Figure 2 ;

[0063] Figure 5 It is a schematic diagram of the grid processing of the 3D point cloud in space in an embodiment of the present invention;

[0064] Figure 6 It is a schematic diagram of the neural network model structure in an embodiment of the present invention;

[0065] Figure 7 It is a schematic diagram of the person fall detection result in the toilet scenario in an embodiment of the present invention;

[0066] Figure 8 It is a schematic diagram of the person fall detection result in other indoor scenarios in an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0067] Please refer to the appendix Figures 1-8 , to make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. In the following description, numerous specific details are provided for a comprehensive understanding of the present invention. However, those skilled in the art should understand that the present invention can be practiced without these specific details. In other cases, to avoid obscuring the essence of the present invention, the detailed descriptions of well-known structures and devices are omitted.

[0068] The present invention provides a 60G millimeter-wave radar personnel fall detection system and method applied to a smart toilet. As Figure 1 shown, the system of the present invention includes a detection module 1, a signal processing module 2, a coordinate conversion module 3, a grid processing module 4, a neural network processing module 5, a fall determination module 6, and a wireless communication module 7.

[0069] The detection module 1 is installed at the bottom of the toilet in an open-hole embedding manner and includes two 60G millimeter-wave radar modules and a smart toilet seat pressure sensor module. Preferably, as Figure 2 shown, two sets of radar detection systems are installed inside the smart toilet, with a height of 0.3 meters, a pitch angle of 15°, a horizontal angle of 30°, a horizontal field of view (FOV) of ±60°, and a coverage range of 3 meters. The open-hole embedding installation method can ensure the accuracy of radar detection while not affecting the aesthetics and use of the toilet. In an embodiment of the present invention, the radar system installed on the left side is the host, which is responsible for receiving the fall detection information of the slave radar and jointly controlling the Bluetooth module with the ESP32 module to transmit the data to the mobile phone APP or the computer-side upper computer.

[0070] The signal processing module 2 is connected to the detection module 1 and is used to process the echo data of multiple chirp signals collected by the two 60G millimeter-wave radar modules. In addition, in this embodiment, the millimeter-wave radar emits high-frequency chirp signals and receives the reflected echo signals. These echo signals contain information such as the distance and speed of the target object. Specifically, as Figure 3 shown, the signal processing module 2 processes the received signals, extracts the distance and speed information of the target by performing discrete Fourier transforms in the fast time dimension (distance dimension) and the slow time dimension (speed dimension). Then, the angle information (horizontal angle and pitch angle) of the target is calculated through beam synthesis technology, and finally, the three-dimensional point cloud data of the target is generated, reflecting the position and dynamic characteristics of the target in the three-dimensional space.

[0071] In an embodiment of the present invention, the echo intermediate-frequency signal of the 60G millimeter-wave radar can be expressed as:

[0072]

[0073] Among them, is the echo intermediate frequency signal of the k-th chirp signal; A is the signal amplitude, usually related to the radar cross section and distance of the target; j is the imaginary unit, representing is the difference frequency, related to the target distance; t is the time variable, in seconds; f d is the Doppler frequency, related to the target velocity; t c is the duration of each chirp, in seconds; k is the chirp signal index, representing which chirp signal.

[0074] Difference frequency can be calculated by the following formula:

[0075]

[0076] Among them, μ is the frequency modulation slope, representing the chirp signal frequency change rate, in Hz / s; τ k is the signal round-trip time delay, in seconds; R is the distance from the target to the radar, in meters; c is the speed of light, approximately 3×10 8 m / s. By performing an FFT operation on the fast time dimension of , the difference frequency can be obtained and the target distance R can be calculated.

[0077] Doppler frequency f d can be calculated by the following formula:

[0078]

[0079] Among them, v is the radial velocity of the target relative to the radar, in m / s; f c is the carrier frequency of the radar, 60 GHz in the present invention; c is the speed of light. By performing an FFT operation on the slow time dimension of , the Doppler frequency f d can be obtained and the target velocity v can be calculated.

[0080] Furthermore, the signal processing module 2 uses the echo signal and combines the antenna distribution of the radar to construct the MVDR power spectrum in each direction, and obtains the horizontal and pitch angles of the current target by searching. The calculation method is as follows:

[0081]

[0082] Among them, P MVDR(θ, Φ) is the MVDR (Minimum Variance Distortionless Response) power spectrum in the direction (θ, Φ); a(θ, φ) is the steering vector of the radar, representing the response when the signal arrives at the antenna array from the direction (θ, Φ); a H (θ, Φ) is the conjugate transpose of the steering vector; R is the signal covariance matrix; R -1 is the inverse matrix of the covariance matrix; θ is the horizontal angle, usually in the range of [-90°, 90°]; Φ is the elevation angle, usually in the range of [-90°, 90°].

[0083] The calculation formula for the signal covariance matrix R is:

[0084]

[0085] where E{·} is the expectation operator, representing the statistical average of the signal; S IF is the radar received signal matrix, and each row corresponds to the received signal of an antenna channel; is the conjugate transpose of S IF . In actual calculation, the average of multiple time samples is usually used to approximate the expectation operation.

[0086] The expression for the steering vector a(θ, Φ) is:

[0087]

[0088] where d x is the antenna spacing of the radar antenna in the x-axis direction, which is set to 2.5 mm in the present invention; λ is the radar operating wavelength. For a 60 GHz radar, M is the number of antenna elements, which is 4 in the present invention; the superscript T represents the transpose operation of the vector.

[0089] By searching for the angles that make P MVDR (θ, Φ) take the maximum value, the direction of the target can be obtained:

[0090]

[0091] where θ target is the horizontal angle of the target; Φ target is the elevation angle of the target; argmax is the operation of taking the independent variable that makes the function take the maximum value. In actual implementation, the grid search method is usually adopted to calculate P MVDR (θ, φ) at a certain angle step (such as 1°) within the value ranges of θ and Φ, and find the angle pair that makes it take the maximum value.

[0092] The coordinate transformation module 3 is connected to the signal processing module 2, and is used to transform the point cloud target in the radar coordinate system into the world coordinate system and map these point clouds to a specific area in the three-dimensional space. Preferably, the coordinate transformation is calculated using the following formula:

[0093]

[0094] where (x w , y w , z w ) are the Cartesian coordinates of the point cloud in the world coordinate system, with the unit of meter; R is the rotation matrix, representing the rotation relationship from the radar coordinate system to the world coordinate system; (r, θ, Φ) are the spherical coordinates of the point cloud in the radar coordinate system, where r is the distance (unit: meter), θ is the horizontal angle (unit: radian), and Φ is the pitch angle (unit: radian); (x offset , y offset , z offset ) is the position offset of the radar in the world coordinate system, with the unit of meter.

[0095] The rotation matrix R is composed of three basic rotation matrices:

[0096]

[0097] where R x (α) is the rotation matrix for rotating α degrees around the x-axis; R y (β) is the rotation matrix for rotating β degrees around the y-axis; R z (γ) is the rotation matrix for rotating γ degrees around the z-axis; α is the rotation angle of the radar around the x-axis, which is set to 15° (pitch inclination angle) in the present invention; β is the rotation angle of the radar around the y-axis, which is usually set to 0° in the present invention; γ is the rotation angle of the radar around the z-axis, which is set to 30° (horizontal inclination angle) in the present invention. These angles need to be converted to radians for calculation.

[0098] In a specific embodiment of the present invention, the installation height of the radar is 0.3 meters, so z offset = 0.3 meters, the pitch inclination angle is 15°, that is, α = 15° × π / 180 = 0.2618 radians, the horizontal inclination angle is 30°, that is, γ = 30° × π / 180 = 0.5236 radians, and β = 0 radians. Through such coordinate transformation, the point cloud data detected by the radar can be accurately mapped to the actual three-dimensional space, providing reliable spatial information for subsequent fall detection.

[0099] The grid processing module 4 is connected to the coordinate transformation module 3, and is used to perform grid processing on the three-dimensional space data, divide the space into multiple cells, and use the number of point clouds as the value of the space unit, and arrange these values as a column vector according to the spatial position. In an embodiment of the present invention, such asFigure 5 As shown, the grid processing steps are as follows:

[0100] First, based on the minimum and maximum coordinate values of the point cloud data, determine the spatial range [x min , x max , [y min , y max , [z min , z max of the target area. In this embodiment, the typical spatial range is [-2m, 2m], [-2m, 2m], [-0.5m, 2m]. This range is determined based on the size of the actual bathroom and the human activity space.

[0101] Then, evenly divide the determined spatial range into small cube grids of the same size, and the size of each grid unit can be set according to the required resolution. In this embodiment, the grid size for spatial division is:

[0102]

[0103] where Δ x , Δ y , Δ z are the grid sizes in the x, y, and z directions respectively, with the unit of meter; M is the number of grid divisions. Preferably, M is set to 20, so there will be 20 grids in each direction, forming a total of 20×20×20 = 8000 grid units. Selecting 20 as the number of grids is based on experimental results, achieving a good balance between computational complexity and detection accuracy.

[0104] Next, for each point in the point cloud data, find which grid unit it belongs to according to its coordinates (x, y, z). Specifically, the grid index is calculated as follows:

[0105]

[0106] where i, j, and k are the grid indices of the point in the x, y, and z directions respectively; represents the floor operation to ensure that the index is an integer. When the point cloud coordinates exceed the defined spatial range, the point can be either ignored or classified into the nearest boundary grid.

[0107] For each grid unit, count the number of points belonging to this grid, and at the same time count and calculate the signal-to-noise ratio, average reflection intensity, center position, etc. of these points, providing useful information for subsequent processing. In this embodiment, mainly count the number of points as the characteristic value of the grid because the density distribution of points is an important feature for distinguishing human postures.

[0108] Finally, rearrange these grid data into a spatial vector G. Preferably, in a linear indexing manner in the order of z, y, x, flatten the three-dimensional grid data into a one-dimensional vector, and the vector length is h×M×M. Specifically, the formula for converting the three-dimensional index (i, j, k) to the one-dimensional index l is

[0109] l = i + j×M + k×M×M,

[0110] In this way, the three-dimensional spatial information is encoded into a one-dimensional vector, which is convenient for subsequent processing by the neural network. Each element in the vector G corresponds to the number of point clouds in a grid cell, reflecting the point cloud density at that spatial position.

[0111] The neural network processing module 5 is connected to the grid processing module 4, and is used to process and train the grid by using the spatial vector value as the input of the fully connected neural network to obtain network weights, and deploy the trained network to the processor. In an embodiment of the present invention, as Figure 6 shown, a three-layer fully connected neural network structure is adopted, including an input layer, two hidden layers and an output layer. The forward calculation process of the neural network is carried out according to the following formula:

[0112] f(G) = σ(W3·ReLU(W2·ReLU(W1·G + b1)+b2)),

[0113] where f(G) is the output of the neural network, representing the probability of falling, and the value range is [0,1]; G is the input grid spatial vector, and the dimension is M 3 (8000 in this embodiment); W1 is the weight matrix of the first fully connected layer, and the dimension is 256×M 3 ; W2 is the weight matrix of the second fully connected layer, and the dimension is 64×256; W3 is the weight matrix of the output layer, and the dimension is 1×64; b1 is the bias vector of the first layer, and the dimension is 256; b2 is the bias vector of the second layer, and the dimension is 64; ReLU(·) is the rectified linear unit activation function, defined as ReLU(x) = max(0,x), and applied element-wise to the vector; σ(·) is the Sigmoid activation function, defined as Mapping the output to the interval (0,1) represents the probability

[0114] The training process of the neural network uses the cross-entropy loss function:

[0115]

[0116] where L is the cross-entropy loss value, measuring the difference between the model prediction and the true label; N is the number of samples in the training batch; y i is the true label of the i-th sample, and the value is 0 (not falling) or 1 (falling); is the predicted probability of the i-th sample, that is, the fall probability f(G i ) output by the model for this sample. The cross-entropy loss function is applicable to binary classification problems and can effectively guide the model to learn to distinguish between fall and non-fall states.

[0117] In a specific embodiment of the present invention, the dimension of the input vector G is 8000 (corresponding to the number of grids of 20×20×20), the number of neurons in the first hidden layer is 256, the number of neurons in the second hidden layer is 64, and the output layer has 1 neuron, representing the probability of falling. The training uses the Adam optimizer, the learning rate is set to 0.001, the batch size is 32, and the number of training epochs is 100 rounds. The selection of these hyperparameters is the best configuration based on a large number of experiments, achieving a good balance between the model convergence speed and performance. The training dataset includes fall and non-fall samples in various scenarios, with a total sample size of approximately 2000, and the ratio of fall samples to non-fall samples is approximately 1:3 to reflect the relatively rare occurrence of fall events in actual applications. After training, the neural network model is deployed on the ESP32 chip for real-time calculation of the fall probability.

[0118] The fall determination module 6 is connected to the neural network processing module 5 and the intelligent toilet seat pressure sensor module, and is used to determine the fall status of the person by combining the fall probability value output by the fully connected neural network and the data of the intelligent toilet seat pressure sensor. In an embodiment of the present invention, the fall determination steps are as follows:

[0119] First, obtain the fall probability value P radar output by the fully connected network pressure and the pressure sensor value P radar of the intelligent toilet seat. The closer P radar is to 1, the more certain the model is that a fall event has been detected; the closer P p is to 0, the more certain the model is that no fall event has been detected. The threshold of the pressure sensor is set to T p ressure. If P pressure ressure>T pressure , it means that someone is touching the toilet seat. In this embodiment, T

[0120] is set to 50N because when an adult slightly touches the toilet seat, it usually generates a pressure of about 30 - 70N. Setting it to 50N can effectively distinguish the state of someone sitting on the toilet from the state of no one. radar and w pressure , and calculate the fusion probability:

[0121]

[0122] Among them, P combined is the fused fall probability, and its value range is [0, 1]; w radar is the weight of the radar detection result, indicating the importance of radar data in the fusion process; P radar is the fall probability detected by the radar, and its value range is [0, 1]; w pressure is the weight of the pressure sensor, indicating the importance of pressure data in the fusion process; P pressure is the measured value of the pressure sensor, with the unit of Newton (N); P max is the maximum range of the pressure sensor, which is used to normalize the pressure value to the [0, 1] interval. In this embodiment, it is set to 200 N, which is considered in view of the adult body weight range and the maximum possible pressure applied.

[0123] Preferably, the weight parameter w radar is set to 0.7, w pressure is set to 0.3, and it satisfies w radar +w pressure = 1. Such a weight allocation is the optimization result based on a large amount of experimental data, considering the main contribution of the millimeter-wave radar to fall detection and the auxiliary verification role of the pressure sensor. The radar data has a higher weight because the radar can directly observe the changes in human body postures, while the pressure sensor mainly provides auxiliary information, such as confirming whether there is someone near the toilet.

[0124] Finally, the fall state is determined according to the fused probability value: if P combined > 0.8, it means that the user has fallen; if P combined ≤ 0.5, it means that the target has not fallen; if 0.8 ≥ P combined > 0.5, it means an undetermined state, and it is necessary to further combine multi-frame data for re-determination. The selection of the thresholds 0.8 and 0.5 is based on the trade-off between the false alarm rate and the missed alarm rate, ensuring that the system can not only detect fall events in a timely manner but also not generate frequent false alarms. Specifically, the high threshold of 0.8 ensures that the alarm is triggered only when the system is very certain that a fall is detected, reducing false alarms; the low threshold of 0.5 serves as the dividing line between possible falls and non-fall states. When the probability is between 0.5 - 0.8, the system will further observe and collect data before making a judgment.

[0125] In the undetermined state, the system will continuously monitor 5 - 10 frames of data (about 1 - 2 seconds). If more than 60% of the frames are determined to be falls (P combined > 0.8) during this period, then it is finally determined to be in the fall state; otherwise, it is determined to be in the non-fall state. This multi-frame determination strategy can effectively reduce false alarms caused by instantaneous interference or posture changes.

[0126] The wireless communication module 7 is connected to the fall determination module 6 and is used to send the determination result of the fall detection system to the mobile phone APP terminal for result display. In an embodiment of the present invention, the wireless communication module adopts Bluetooth 4.2 technology, and the transmission distance can reach 10 meters, which is sufficient to cover the home environment. The power consumption of the Bluetooth module is relatively low, about 10 - 30 mW, which is suitable for a monitoring system with long-term operation. Through the Bluetooth module, the system can send the fall detection result to the user's mobile phone APP in real time, providing timely alarm and monitoring services. The data transmission frequency is 2 times per second, and an alarm message will be sent immediately when a fall event is detected to ensure timely response.

[0127] The present invention also provides a 60G millimeter-wave radar personnel fall detection method applied to a smart toilet. This method corresponds to the above system and includes the following steps:

[0128] Step 1: Install two 60G millimeter-wave radar modules at the bottom of the toilet in an open-hole embedded installation manner.

[0129] Step 2: Obtain the echo data of multiple chirp signals through the two 60G millimeter-wave radar modules, perform discrete Fourier transform on them in the fast time dimension and the slow time dimension, and use digital beamforming technology to obtain the horizontal angle and elevation angle of the target, generating the three-dimensional point cloud information of the target.

[0130] Step 3: Convert the point cloud target in the radar coordinate system to the world coordinate system, and map these point clouds to a specific area in the three-dimensional space.

[0131] Step 4: Perform grid processing on the mapped three-dimensional space data, divide the space into multiple cells, and use the number of point clouds as the value of the space unit, arranging them as a column vector according to the spatial position.

[0132] Step 5: Use the spatial vector value as the input of a fully connected neural network for processing and training to obtain the weights of the network.

[0133] Step 6: Deploy the trained network to the ESP32 processor, and use the spatial vector generated in real time by the millimeter-wave radar as the input of the network to calculate the probability value of a person falling.

[0134] Step 7: Combine the fall probability value with the pressure sensor data of the smart toilet seat to determine the fall situation of the person.

[0135] Step 8: Send the determination result of the fall detection system to the mobile phone APP terminal through the Bluetooth module for result display.

[0136] In a practical application embodiment of the present invention, such as Figure 7 and Figure 8As shown, the system can accurately identify the fall status of personnel in different scenarios. Figure 7 The fall detection results in the toilet scenario are shown. The left side shows the distribution of point cloud data, and the right side is the corresponding actual scenario image. Figure 8 The fall detection results in other indoor scenarios are shown. From these results, it can be seen that the system provided by the present invention can effectively distinguish normal activities (such as sitting down, standing up, walking, etc.) and fall status, and has good detection accuracy and environmental adaptability.

[0137] In another embodiment of the present invention, the system can adapt to different usage scenarios and user requirements by adjusting the structure and parameters of the neural network. For example, for elderly care institutions, the depth and complexity of the network can be increased to improve the detection accuracy; while for home use, the network structure can be simplified to reduce the computational complexity and extend the battery life of the device. In addition, the system can further improve the detection accuracy and reliability by adding other types of sensors (such as temperature sensors, humidity sensors, etc.).

[0138] In summary, the 60G millimeter-wave radar personnel fall detection system and method for intelligent toilets provided by the present invention achieve high-precision and non-contact personnel fall detection through multi-sensor fusion and deep learning technologies. This system has the advantages of privacy protection, strong environmental adaptability, and no need to wear devices, solves many problems existing in traditional fall detection technologies, and provides an innovative technical solution for the fields of smart home and health monitoring.

[0139] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. The 60G millimeter wave radar personnel fall detection system applied to smart toilets is characterized by: The system comprises: The detection module is installed at the bottom of the toilet by opening and embedding, and includes two 60G millimeter-wave radar modules and a smart toilet seat pressure sensor module; A signal processing module, connected to the detection module, for processing the echo data of multiple chirp signals collected by the two 60G millimeter wave radar modules, wherein the processing includes performing discrete Fourier transform in fast time dimension and slow time dimension, and using digital beam synthesis technology to obtain the horizontal angle and pitch angle of the target, and generating three-dimensional point cloud information of the target; A coordinate conversion module, connected to the signal processing module, for converting point cloud targets in the radar coordinate system to the world coordinate system, and mapping these point clouds to specific areas in the three-dimensional space; A grid processing module, connected to the coordinate conversion module, is used to perform grid processing on the three-dimensional spatial data, divide the space into a plurality of cells, and use the number of point clouds as the value of the spatial unit, and arrange these values ​​into column vectors according to the spatial position; A neural network processing module, connected to the grid processing module, is used to process and train the grid using the space vector value as the input of the fully connected neural network to obtain network weights, and deploy the trained network to the processor; A fall determination module, connected to the neural network processing module and the smart toilet seat pressure sensor module, for determining a person's fall condition by combining the fall probability value output by the fully connected neural network and the data of the smart toilet seat pressure sensor; The wireless communication module is connected to the fall determination module and is used to send the determination result of the fall detection system to the mobile phone APP terminal for displaying the result.

2. According to claim 1, the 60G millimeter wave radar personnel fall detection system applied to the smart toilet is characterized by: The detection module is installed on the outside of the smart toilet by opening and embedding.

3. The 60G millimeter wave radar personnel fall detection system applied to the smart toilet according to claim 1 is characterized in that: The coordinate conversion module converts the radar coordinate system to the world coordinate system according to the following formula: Among them, (r, θ, Φ) represents the polar coordinates of the target point cloud in the radar coordinate system, where r represents the distance, θ represents the horizontal angle, and Φ represents the pitch angle; (x w ,y w ,z w ) represents the coordinate value of the target point cloud in the world coordinate system; Indicates the rotation angle of the radar around the x, y, and z axes; x offset y offset , z offset Indicates the displacement of the radar on the x, y, and z axes.

4. The 60G millimeter wave radar personnel fall detection system applied to a smart toilet according to claim 1, characterized in that: The grid processing module is performed according to the following steps: Based on the 3D point cloud data collected by the millimeter wave radar, find the minimum and maximum coordinate values ​​of all points and determine the range of the target area in the x, y and z directions; The determined spatial range is evenly divided into small cubic grids of the same size. The size of each grid unit can be set according to the required resolution; For each point in the point cloud data, find which grid cell it belongs to based on its x, y, and z coordinate values; For each grid cell, count the number of points belonging to the grid, and at the same time count and calculate the signal-to-noise ratio, average reflection intensity, center position, etc. of these points to provide useful information for subsequent processing; Outputs grid data in vector form, where each grid contains its spatial position and statistical information.

5. The 60G millimeter wave radar personnel fall detection system applied to the smart toilet according to claim 1 is characterized in that: The fully connected neural network in the neural network processing module is trained according to the following formula: f(G)=σ(W3·ReLU(W2·ReLU(W1·G+b1)+b2)), Where W1, W2 and W3 are weight matrices of the fully connected layer, b1 and b2 are bias terms, σ(·) represents the activation function, and the output value f(G) is the probability of falling. L is the cross entropy loss value during network training.

6. The 60G millimeter wave radar personnel fall detection system applied to the smart toilet according to claim 1 is characterized in that: The fall determination module determines the fall status of a person according to the following steps: Get the fall probability value P output by the fully connected network radar And the pressure sensor value P of the smart toilet seat pressure ; Define the threshold value T of the pressure value pressure , if P pressure >T pressure , indicating that someone has touched the toilet seat; Define the weight parameter w radar and w pressure , and calculate the fusion probability If P combined ≥T combined , it is judged as a fall state, and the system sends the result to the mobile phone APP for display through the wireless communication module.

7. The 60G millimeter wave radar personnel fall detection system applied to the smart toilet according to claim 1 is characterized in that: The processor is an ESP32 chip.

8. The 60G millimeter wave radar personnel fall detection system applied to a smart toilet according to claim 6, characterized in that: The w radar Set to 0.7, the w pressure Set to 0.3, the P combined >0.8 indicates that the user falls. combined ≤0.5 means the target did not fall, 0.8≤P combined <0.5 indicates an undetermined state, which requires further determination based on multiple frames of data.

9. The 60G millimeter wave radar personnel fall detection system applied to the smart toilet according to claim 1, characterized in that: The wireless communication module is a Bluetooth module.

10. The 60G millimeter wave radar personnel fall detection method applied to a smart toilet adopts any one of claims 1 to 9, characterized in that: The method comprises the following steps: Two 60G millimeter-wave radar modules are installed at the bottom of the toilet by opening and embedding. The two 60G millimeter-wave radar modules are used to obtain echo data of multiple chirp signals, and discrete Fourier transform is performed on them in the fast time dimension and the slow time dimension. The digital beam synthesis technology is used to obtain the horizontal angle and the pitch angle of the target, and the three-dimensional point cloud information of the target is generated; Convert the point cloud targets in the radar coordinate system to the world coordinate system and map these point clouds to specific areas in three-dimensional space; The mapped three-dimensional spatial data is gridded to divide the space into multiple cells, and the number of point clouds is used as the value of the spatial unit, which is arranged as a column vector according to the spatial position; The spatial vector value is used as an input of a fully connected neural network for processing and training to obtain a weight of the network; The trained network is deployed to the ESP32 processor, and the millimeter-wave radar generates spatial vectors in real time as the network input to calculate the probability of a person falling. Combining the fall probability value with the pressure sensor data of the smart toilet seat, determining the fall condition of the person; The judgment results of the fall detection system are sent to the mobile phone APP terminal through the Bluetooth module for displaying the results.

Images