A fall detection method and system based on millimeter wave radar

Abstract

The present invention relates to a fall detection method and system based on millimeter-wave radar. The detection method comprises: S1, acquiring three-dimensional point cloud data detected by the millimeter-wave radar; S2, converting the three-dimensional point cloud data into the position and speed data of each target, and bringing the data into the Kalman cycle prediction to achieve a prediction for the next moment, matching the position of the target at the current moment with the predicted position of the target at the previous moment, and obtaining the ID of each target at the current moment; S3, performing fall detection and judgment based on the ID, position, and speed information of the target. The present application uses a Kalman filter to predict the state of a target at a certain moment, and further confirms the actual action of a low-speed fall by setting a speed threshold in the three-dimensional direction of the target.

Description

Technical Field

[0001] The present invention belongs to the field of target recognition technology, and specifically relates to a fall detection method and system based on millimeter-wave radar. Background Art

[0002] A fall is the accidental collapse of a person to the ground or a low position, and it is one of the most common abnormal human movements. According to the World Health Organization, falls are the second leading cause of accidental injury and death worldwide. The health risks associated with falls increase exponentially with age. For the elderly, falls can cause serious consequences such as cuts, abrasions, or fractures. The psychological damage caused by falls cannot be ignored either. Many elderly people have developed a fear of falling, which affects their quality of life. However, the uncertainty of the timing of falls and the difficulty of preventing them pose practical difficulties for rescuing elderly people after falls. Therefore, timely detection of falls in the elderly has attracted the attention of many researchers. In addition, the slow response to calls for help after a fall is also a major factor in the reduced life expectancy of the elderly.

[0003] The aging population is driving the development of new healthcare services, evolving from traditional hospital-based treatment to home-based care services, where patients, leveraging emerging technologies, can receive a portion of their treatment at home. This family-centered approach to long-term care not only improves patients' quality of life but also indirectly reduces healthcare costs for both patients and society. Elderly individuals living alone, whether due to home healthcare or other reasons, are prone to falls. To ensure the safety of elderly individuals living independently, it is essential to promptly detect falls so that medical staff or family members can be notified in an emergency.

[0004] In their study on fall detection methods based on millimeter-wave radar, Sun Jizhou et al. proposed a fall detection method based on millimeter-wave radar. This method uses the millimeter-wave radar to obtain target position information and performs mean filtering on the position coordinate sequence. Then, through piecewise linear fitting, it approximates the motion pattern of the monitored object. This method comprehensively considers the vertical and horizontal position changes and the corresponding relationship between the two when a person falls. After detecting a fall, the fall mode is identified based on multiple thresholds. If a fall is detected, an alarm is immediately issued. However, the method described in this article is implemented in an open area and does not address collisions with other objects in the environment. It also does not consider fall detection in the presence of multiple people.

[0005] A Chinese invention patent with publication number CN112782664A, published on May 11, 2021, discloses a bathroom fall detection method based on millimeter-wave radar. Specifically, the method uses deployed millimeter-wave radars to collect user data; performs coordinate conversion on the user data to determine its three-dimensional point cloud data in the actual scene; removes static objects to obtain a point cloud map of dynamic objects, and removes interference noise; and finally, performs fall detection based on a comprehensive judgment of the user's aspect ratio greater than a certain threshold, the highest point less than a certain threshold, and the descent speed greater than a certain threshold. However, not all falls are fast movements, and some non-falling movements also have high velocities. This is because in many cases, people grab onto furniture or lean against a wall when falling, which slows down the fall and causes the rate to decrease. It is also common for elderly people to fall from chairs or wheelchairs. In such scenarios, the falling action reduces the speed and high-frequency energy value. It also cannot distinguish between falls caused by one person in a group.

[0006] Therefore, how to extract appropriate features from the various data collected by millimeter-wave radar and perform accurate fall detection is an area worthy of research. Therefore, a high-precision detection method is urgently needed that can monitor people's movements in real time, provide fall alarms, enable rapid rescue, and minimize injuries. Furthermore, how to detect falls in multiple people is an unresolved issue. Summary of the Invention

[0007] To overcome the problems in the background technology, this patent provides a design solution for fall detection based on millimeter-wave radar, which can quickly detect people who fall when there are multiple targets; and can also accurately detect fall events when the fall speed is not fast enough.

[0008] The present invention proposes a fall detection method based on millimeter wave radar, which includes:

[0009] S1, obtains the three-dimensional point cloud data detected by the millimeter wave radar;

[0010] S2, converting the three-dimensional point cloud data into the position and velocity data of each target, and bringing it into the Kalman cycle prediction to realize the prediction of the next moment, matching the position of the target at the current moment with the predicted position of the target at the previous moment, and obtaining the ID of each target at the current moment;

[0011] S3, performing fall detection and judgment based on the ID, position and speed information of the target.

[0012] Furthermore, S2 is specifically:

[0013] S21, first calculate the initial observation value, set x1, y1, z1 as the coordinates of the current target at the first moment in the x, y, z direction, vx1, vy1, vz1 as the speed of the current target at the first moment in the x, y, z direction, x k ,y k ,z k is the coordinate of the current target at the kth moment in the x, y, and z directions, vx k 、vy k 、vz k is the velocity of the current target in the x, y, and z directions at the kth moment, where k is a variable, k = 2, 3, ..., N, and N represents the total time of target observation;

[0014] is the observed value at the k-1th moment, then Expressed as: Among them, vx k-1 、vy k-1 、vz k-1 Through the target's position x at the k-1th moment k-1 ,y k-1 , z k-1 and the position x at the k-2th moment k-2 ,y k-2 , z k-2 calculated;

[0015]

[0016] S22, set the estimated difference of each observation value, that is e k-1 is a 6*1 matrix

[0017] S23, according to e k-1 Calculate the prior estimate covariance matrix P at time k-1 k-1 =e k-1 e k-1 T , P k-1 It is a 6*6 matrix;

[0018] S24, calculate the observation vector and P k-1 Bring in the Kalman cycle prediction to realize the prediction of the next moment.

[0019] Furthermore, the Kalman cycle prediction is specifically as follows:

[0020] 1) Set the initial P k-1 as well as is the prior prediction of the observation value at time k, A is the state transfer matrix, Bμ k-1is the state control quantity, usually zero, is the posterior estimated covariance matrix, P k-1 is the prior estimated covariance matrix, Q is the process excitation noise covariance matrix;

[0021] 2) Forward calculation of state variables

[0022] 3) Forward error covariance

[0023] 4) Calculate Kalman gain

[0024] 5) From the observed variable z k Update Estimates Get the predicted value at time k;

[0025] 6) Update the prior estimated covariance matrix at time k

[0026] 7) When the point cloud data at the kth moment is newly acquired, the target position detected at the kth moment is compared with the target position predicted at the kth moment. When the Euclidean distance is less than the threshold th0, it is represented as the same target. Otherwise, it is a new target. The index value of the target is increased by 1, and i is used to represent the index value of the target. The x, y, and z coordinates of the i-th target at the k-th moment are x k [i],y k [i],z k [i] represents speed, and vx represents speed. k [i],vy k [i],vz k [i] indicates;

[0027] Among them, th0 is the distance threshold between the target position detected at time k and the predicted position;

[0028] Repeat steps 2)-7) to complete the prediction and matching at each subsequent moment.

[0029] Further, S3 is specifically as follows: S31, obtain the x, y, and z coordinates x of the i-th target at the k-th moment according to the method in step S2 k [i],y k [i],z k [i], and speed vx k [i],vy k [i],vz k [i];

[0030] S32, calculate the speed in the xy direction at the kth moment

[0031] S33, when the height of the i-th target, i.e., the z coordinate, is lower than the set first height threshold th1, it means that the height has dropped, and the process proceeds to step S34;

[0032] S34, the cumulative value of the number of altitude drops fall_count[i] is increased by 1;

[0033] S35, use FALL[i] to represent the height descent state of the i-th target, and FALL[i] is initialized to 0. When the height descent state FALL[i] = 0 and the cumulative value of the height descent count fall_count[i] is increased by 1, the height descent state FALL[i] is set to 1. This moment is the time m when the height descent occurs. Record the x and y coordinates at time m-1, which is called the horizontal position before the height descent, recorded as x m-1 [i],y m-1 [i], otherwise go directly to S36;

[0034] S36, calculate the maximum difference dxy between the x, y coordinate position of the target and the horizontal position before the height drop at each subsequent moment k [i];

[0035]

[0036] S37, when FALL[i]=1, when the falling speed in the z direction at the kth moment is large enough and the speed change in the xy direction is large enough, that is, vz k [i]<th2&&vxy k [i]>th3, it means a rapid fall has occurred, then FALL[i]=2, where th2 is the second height threshold and th3 is the third height threshold; otherwise, when the speed of descent in the z direction at the kth moment is generally large and the speed change in the xy direction is generally large, that is, vz k [i]<th4&&vxy k When [i]>th5, it indicates a slow fall has occurred, FALL[i]=3; otherwise, i=i+1, and the process returns to step S33 to continue judging the i+1th target, where th4 is the fourth height threshold and th5 is the fifth height threshold;

[0037] S38, when the number of falling frames is greater than th6, enter S39, th6 is the threshold of the number of falling frames, otherwise i=i+1, return to S31;

[0038] S39, when FALL[i]=2, it is determined to be a fall, otherwise, enter S310;

[0039] S310, when FALL[i]=3 and dxy kWhen [i]>th7, it is judged as a fall, otherwise, i=i+1, and the process returns to S31; wherein th7 is the threshold for rapid sitting down.

[0040] Furthermore, the threshold value of the number of falling frames th6=5.

[0041] Furthermore, in S33 , th1 is a height threshold, indicating the height value after a fall occurs, and the value range of th1 is 0.3 m < th1 < 1 m.

[0042] Furthermore, the set duration described in S35 is 3 frames.

[0043] The present application also provides a fall detection system based on millimeter wave radar, the detection system comprising:

[0044] A data acquisition module, wherein the data acquisition module is used to acquire three-dimensional point cloud data detected by the millimeter wave radar;

[0045] A Kalman filter module converts the three-dimensional point cloud data into the position and velocity data of each target, and feeds the data into the Kalman loop prediction to achieve the prediction of the next moment. The ID of each target at the current moment is obtained by matching the position of the target at the current moment with the predicted position of the target at the previous moment;

[0046] A fall detection module performs fall detection based on the target's ID, location, and speed information.

[0047] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0048] The millimeter-wave radar-based fall detection solution provided in this application uses only millimeter-wave radar, a non-contact monitoring technology, to promptly detect falls. This monitoring method does not involve any sound, image, or video information, thus avoiding potential privacy leaks and providing an excellent option for privacy-conscious users. It also effectively addresses the safety concerns of unsupervised elderly people, resulting in significant economic and social benefits.

[0049] This application uses the Kalman filter to predict the state of the target at a certain moment;

[0050] The real action of falling at low speed is further confirmed by setting the speed threshold of the target in the three-dimensional direction. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flow chart of the millimeter-wave radar-based fall detection method of the present invention;

[0052] Figure 2This is a flowchart of the millimeter-wave radar-based fall detection method of the present invention. DETAILED DESCRIPTION

[0053] The technical solution of the present invention is further described in detail below with reference to the accompanying drawings:

[0054] Example 1

[0055] The present invention proposes a fall detection method based on millimeter wave radar, the detection method comprising:

[0056] S1, obtains the three-dimensional point cloud data detected by the millimeter wave radar;

[0057] 1) The millimeter-wave radar returns coordinates in the polar coordinate system. The coordinates of each point cloud in the polar coordinate system are Among them, n represents the nth point cloud.

[0058] 2) Converted to the world coordinate system, the coordinates of each point cloud in the world coordinate system are (x n ,y n , z n )

[0059] z n =r n *sin(θ n )

[0060]

[0061]

[0062] S2, converting the three-dimensional point cloud data into the position and velocity data of each target, and bringing it into the Kalman cycle prediction to realize the prediction of the next moment, matching the position of the target at the current moment with the predicted position of the target at the previous moment, and obtaining the ID of each target at the current moment;

[0063] S3, performing fall detection and judgment based on the ID, position and speed information of the target.

[0064] When a fall occurs, an alarm is triggered to alert children. After a fall, when the z-coordinates of all target IDs are greater than a set value, it indicates that standing up has occurred.

[0065] Furthermore, S2 is specifically:

[0066] S21, first calculate the initial observation value, set x1, y1, z1 as the coordinates of the current target at the first moment in the x, y, z direction, vx1, vy1, vz1 as the speed of the current target at the first moment in the x, y, z direction, x k ,y k ,zk is the coordinate of the current target at the kth moment in the x, y, and z directions, vx k 、vy k 、vz k is the velocity of the current target in the x, y, and z directions at the kth moment, where k is a variable, k = 2, 3, ..., N, and N represents the total time of target observation;

[0067] is the observed value at the k-1th moment, then Expressed as: Among them, vx k-1 、vy k-1 、vz k-1 Through the target's position x at the k-1th moment k-1 ,y k-1 , z k-1 and the position x at the k-2th moment k-2 ,y k-2 , z k-2 calculated;

[0068]

[0069] S22, set the estimated difference of each observation value, that is e k-1 is a 6*1 matrix;

[0070] S23, according to e k-1 Calculate the prior estimate covariance matrix P at time k-1 k-1 =e k-1 e k-1 T , P k-1 It is a 6*6 matrix;

[0071] S24, calculate the observation vector and P k-1 Bring in the Kalman cycle prediction to realize the prediction of the next moment.

[0072] A typical example of a Kalman filter is predicting the coordinates and velocity of an object from a finite, noisy sequence of observations of its position (which may contain bias). It can start observations at any point and use a recursive filtering approach for calculation. Its characteristic is that it does not require the storage of past measurement data. When new data is measured, new estimates of various quantities are calculated based on the new data and the previous moment's estimated values, using the system's own state transition equations (i.e., dynamic equations) and a set of recursive formulas. It is simple, computationally inefficient, and can be calculated in real time.

[0073] The Kalman filter uses feedback control to estimate the process state: the filter estimates the state of the process at a certain moment and then obtains feedback in the form of (noisy) measurement variables. Therefore, the Kalman filter can be divided into two parts: the time update equation and the measurement update equation. The time update equation is responsible for extrapolating the values ​​of the current state variables and error covariance estimates forward in time to construct a priori estimates for the state at the next time. The measurement update equation is responsible for feedback; that is, it combines the prior estimates with the new measurement variables to construct an improved a posteriori estimate. The time update equation can also be considered an estimation equation, and the measurement update equation can be considered a correction equation.

[0074] The time update equations of the Kalman filter are:

[0075]

[0076]

[0077] in, represents the observation value at time k-1, is the prior prediction of the observation value at time k, A is the state transfer matrix, Bμ k-1 is the state control quantity, usually zero, is the posterior estimated covariance matrix, P k-1 is the prior estimated covariance matrix, and Q is the process excitation noise covariance matrix.

[0078] The measurement update equations of the Kalman filter are:

[0079]

[0080] Among them, M k is the Kalman gain, H is the observation matrix, R is the observation noise covariance matrix, The posterior prediction of the observation at time k, where I is the identity matrix.

[0081] In the present invention, is the six-dimensional observation vector, Among them, x, y, z are the coordinates of the current target, v x ,v y ,v z is the speed of the target in the x-direction, y-direction and z-direction.

[0082] Assuming that the moving target moves in a uniform linear motion within a unit time, the state transfer matrix

[0083] Since the time interval for processing each frame of image is 100 ms, Δt=100 ms.

[0084] Since only x, y, and z can be observed, the observation matrix is

[0085]

[0086] Assume that w and v are both zero-mean and independent noise vectors in all directions, so let their covariance matrix be

[0087]

[0088] Furthermore, the Kalman cycle prediction is specifically as follows:

[0089] 1) Set the initial P k-1 as well as is the prior prediction of the observation value at time k, A is the state transfer matrix, Bμ k-1 is the state control quantity, usually zero, is the posterior estimated covariance matrix, P k-1 is the prior estimated covariance matrix, Q is the process excitation noise covariance matrix;

[0090] 2) Forward calculation of state variables

[0091] 3) Forward error covariance

[0092] 4) Calculate Kalman gain

[0093] 5) From the observed variable z k Update Estimates Get the predicted value at time k;

[0094] 6) Update the prior estimated covariance matrix at time k

[0095] 7) When the point cloud data at the kth moment is newly acquired, the target position detected at the kth moment is compared with the target position predicted at the kth moment. When the Euclidean distance is less than the threshold th0, it is represented as the same target. Otherwise, it is a new target. The index value of the target is increased by 1, and i is used to represent the index value of the target. The x, y, and z coordinates of the i-th target at the k-th moment are x k [i],y k [i],z k [i] represents speed, and vx represents speed. k [i],vy k [i],vz k [i] indicates;

[0096] Where th0 is the distance threshold between the target position detected at time k and the predicted position; the distance threshold between the positions at time k predicted after the Kalman prediction at time k-1 is less than the threshold th0, indicating that they are the same target, otherwise it is a new target;

[0097] Repeat steps 2)-7) to complete the prediction and matching at each subsequent moment.

[0098] Further, S3 is specifically as follows: S31, obtain the x, y, and z coordinates x of the i-th target at the k-th moment according to the method in step S2 k [i],y k [i],z k [i], and speed vx k [i],vy k [i],vz k [i];

[0099] S32, calculate the speed in the xy direction at the kth moment

[0100] S33, when the height of the i-th target, ie, the z coordinate, is lower than the set first height threshold th1, it indicates that the height has dropped, and the process proceeds to step S34; in this embodiment, the set value of th1 is 0.7.

[0101] S34, the cumulative value of the number of altitude drops fall_count[i] is increased by 1;

[0102] S35, use FALL[i] to represent the height descent state of the i-th target, and FALL[i] is initialized to 0. When the height descent state FALL[i] = 0 and the cumulative value of the height descent count fall_count[i] is increased by 1, the height descent state FALL[i] is set to 1. This moment is the time m when the height descent occurs. Record the x and y coordinates at time m-1, which is called the horizontal position before the height descent, recorded as x m-1 [i],y m-1 [i], otherwise directly enter S36; in this embodiment, the duration is set to 3 frames.

[0103] S36, calculate the maximum difference dxy between the x, y coordinate position of the target and the horizontal position before the height drop at each subsequent moment k [i];

[0104]

[0105] S37, when FALL[i]=1, when the falling speed in the z direction at the kth moment is large enough and the speed change in the xy direction is large enough, that is, vz k [i]<th2&&vxy k[i]>th3, indicating a rapid fall, then FALL[i]=2, where th2 is the second height threshold, th3 is the third height threshold. In this embodiment, the setting value of th2 is -1.0, and the setting value of th3 is 1.0; otherwise, when the descent speed in the z direction at the kth moment is generally large and the speed change in the xy direction is generally large, that is, vz k [i] <th4&&vxy k When [i]>th5, it means a slow fall has occurred, FALL[i]=3, otherwise, i=i+1, return to step S33, and continue to judge the i+1th target, where th4 is the fourth height threshold, th5 is the fifth height threshold, the set value of th4 is -0.5, and the set value of th5 is 0.5.

[0106] Among them, th2 is the speed threshold in the z direction when falling rapidly. The setting of this threshold is to detect the speed drop in the z direction, so th2<0, th3 is the speed threshold in the xy direction when falling rapidly, from vxy k From the calculation formula of [i], it can be seen that th3 is a value greater than 0, so th3>0; th4 is the speed threshold in the z direction when falling slowly. This threshold is set to detect the speed drop in the z direction, so th4 is less than 0, 0>th4>th2; th5 is the speed threshold in the xy direction when falling slowly, so the value range is: 0<th5<th3;

[0107] S38, when the number of fall frames is greater than th6, proceed to S39, th6 is the threshold of the number of fall frames, this parameter is closely related to the sampling rate. In the present invention, the sampling rate is 10 frames / s (i.e., 100ms acquisition), so th6 = 5, indicating that the fall process must have lasted for half a second. When the number of fall frames is not greater than th6, i = i + 1, and return to S31;

[0108] S39, when FALL[i]=2, it is determined to be a fall, otherwise, enter S310;

[0109] S310, when FALL[i]=3 and dxy k If [i]>th7, it is determined to be a fall. Otherwise, i=i+1, and the process returns to S31. th7 is the threshold for rapid sitting down. th7 is set to filter out rapid sitting down rather than falls. The set value of th7 is 0.3.

[0110] Furthermore, in S33 , th1 is a height threshold, indicating the height value after a fall occurs, and the value range of th1 is 0.3 m < th1 < 1 m.

[0111] The core invention points of this application are: 1. Predicting the state of a target at a certain moment through a Kalman filter; 2. Further confirming the actual action of a low-speed fall by setting a speed threshold in the three-dimensional direction of the target.

[0112] Example 2

[0113] The present application also provides a fall detection system based on millimeter wave radar, the detection system comprising:

[0114] A data acquisition module, wherein the data acquisition module is used to acquire three-dimensional point cloud data detected by the millimeter wave radar;

[0115] A Kalman filter module converts the three-dimensional point cloud data into the position and velocity data of each target, and feeds the data into the Kalman loop prediction to achieve the prediction of the next moment. The ID of each target at the current moment is obtained by matching the position of the target at the current moment with the predicted position of the target at the previous moment;

[0116] A fall detection module performs fall detection based on the target's ID, location, and speed information.

[0117] This application provides a comprehensive solution for a fall detection system and method based on millimeter-wave radar. Using only millimeter-wave radar, a non-contact monitoring technology, falls can be detected immediately. This monitoring method does not involve any sound, image, or video information, thus avoiding potential privacy leaks and providing an excellent option for privacy-conscious users. It also effectively addresses the safety concerns of unsupervised elderly people, resulting in significant economic and social benefits.

[0118] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, 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 fall detection method based on millimeter wave radar, characterized in that: The detection method comprises: S1, obtains the three-dimensional point cloud data detected by the millimeter wave radar; S2, converting the three-dimensional point cloud data into the position and velocity data of each target, and bringing it into the Kalman cycle prediction to realize the prediction of the next moment, matching the position of the target at the current moment with the predicted position of the target at the previous moment, and obtaining the ID of each target at the current moment; S3, perform fall detection based on the target's ID, location, and speed information; S3 specifically includes: S31, obtain the x, y, and z coordinates of the i-th target at the k-th moment according to the method in step S2 k [i],y k [i],z k [i], and speed vx k [i],vy k [i],vz k [i]; S32, calculate the speed in the xy direction at the kth moment S33, when the height of the i-th target, i.e., the z coordinate, is lower than the set first height threshold th1, it means that the height has dropped, and the process proceeds to step S34; S34, the cumulative value of the number of altitude drops fall_count[i] is increased by 1; S35, use FALL[i] to represent the height descent state of the i-th target, and FALL[i] is initialized to 0. When the height descent state FALL[i] = 0 and the cumulative value of the height descent count fall_count[i] is increased by 1, the height descent state FALL[i] is set to 1. This moment is the time m when the height descent occurs. Record the x, y coordinates at time m-1, which is called the horizontal position before the height descent, recorded as x m-1 [i],y m-1 [i], otherwise go directly to S36; S36, calculate the maximum difference dxy between the x, y coordinate position of the target and the horizontal position before the height drop at each subsequent moment k [i]; S37, when FALL[i]=1, when the falling speed in the z direction at the kth moment is large enough and the speed change in the xy direction is large enough, that is, vz k [i]<th2&&vxy k [i]>th3, indicating a rapid fall, then FALL[i]=2, where th2 is the second height threshold and th3 is the third height threshold; otherwise, when the z-direction descent speed at the kth moment is generally large and the xy-direction speed change is generally large, that is, vz k [i]<th4&&vxy k When [i]>th5, it indicates a slow fall has occurred, FALL[i]=3, otherwise, i=i+1, and the process returns to S33 to continue judging the i+1th target, where th4 is the fourth height threshold and th5 is the fifth height threshold; S38, when the number of falling frames is greater than th6, enter S39, th6 is the threshold of the number of falling frames, otherwise i=i+1, return to S31; S39, when FALL[i]=2, it is determined to be a fall, otherwise, enter S310; S310, when FALL[i]=3 and dxy k When [i]>th7, it is judged as a fall, otherwise, i=i+1, and returns to S31; where th7 is the threshold for rapid sitting down.

2. The fall detection method based on millimeter wave radar according to claim 1, characterized in that: S2 is specifically: S21, first calculate the initial observation value, set x1, y1, z1 as the coordinates of the current target at the first moment in the x, y, z direction, vx1, vy1, vz1 as the speed of the current target at the first moment in the x, y, z direction, x k ,y k ,z k vx is the coordinate of the current target at the kth moment in the x, y, and z directions, k 、vy k 、vz k is the speed of the current target in the x, y, and z directions at the kth moment, where k is a variable. k=2,3,...,N, where N represents the total time of target observation; is the observed value at the k-1th moment, then Expressed as: Among them, vx k-1 、vy k-1 、vz k-1 Through the target's position x at the k-1th moment k-1 ,y k-1 , z k-1 and the position x at the k-2th moment k-2 ,y k-2 , z k-2 calculated; S22, set the estimated difference of each observation value, that is e k-1 It is a 6*1 matrix; S23, according to e k-1 Calculate the prior estimate covariance matrix P at time k-1 k-1 =e k-1 e k-1 T , P k-1 It is a 6*6 matrix; S24, calculate the observation vector and P k-1 Bring in the Kalman cycle prediction to realize the prediction of the next moment.

3. The fall detection method based on millimeter wave radar according to claim 2, characterized in that: The Kalman cycle prediction is specifically: 1) Set the initial P k-1 as well as is the prior prediction of the observation value at time k, A is the state transfer matrix, Bμ k-1 is the state control quantity, usually zero, is the posterior estimated covariance matrix, P k-1 is the prior estimated covariance matrix, Q is the process excitation noise covariance matrix; 2) Forward calculation of state variables 3) Forward error covariance 4) Calculate Kalman gain 5) From the observed variable z k Update Estimates Get the predicted value at time k; 6) Update the prior estimated covariance matrix at time k Among them, M k is the Kalman gain, H is the observation matrix, R is the observation noise covariance matrix, The posterior prediction of the observation value at time k, I is the unit matrix; 7) When the point cloud data at the kth moment is newly acquired, the target position detected at the kth moment is compared with the target position predicted at the kth moment. When the Euclidean distance is less than the threshold th0, it is represented as the same target. Otherwise, it is a new target. The index value of the target is increased by 1, and i is used to represent the index value of the target. The x, y, and z coordinates of the i-th target at the k-th moment are x k [i],y k [i],z k [i] represents speed, and vx represents speed. k [i],vy k [i],vz k [i] indicates; Among them, th0 is the distance threshold between the target position detected at time k and the predicted position; Repeat steps 2)-7) to complete the prediction and matching at each subsequent moment.

4. The fall detection method based on millimeter wave radar according to claim 2, characterized in that: The threshold value of the number of falling frames th6=5.

5. The fall detection method based on millimeter wave radar according to claim 2, characterized in that: In S33 , th1 is a height threshold, indicating the height value after a fall occurs, and the value range of th1 is 0.3 m < th1 < 1 m.

6. The fall detection method based on millimeter wave radar according to claim 2, characterized in that: The set duration in S35 is 3 frames.

7. A fall detection system based on millimeter wave radar, characterized in that: The detection system is used to implement the fall detection method based on millimeter wave radar as claimed in claim 1, and the detection system includes: A data acquisition module, wherein the data acquisition module is used to acquire three-dimensional point cloud data detected by the millimeter wave radar; A Kalman filter module converts the three-dimensional point cloud data into the position and velocity data of each target, and feeds the data into the Kalman loop prediction to achieve the prediction of the next moment. The ID of each target at the current moment is obtained by matching the position of the target at the current moment with the predicted position of the target at the previous moment; A fall detection module performs fall detection based on the target's ID, location, and speed information.

Images