An indoor personnel fall detection method based on millimeter-wave radar

Abstract

The present invention relates to an indoor personnel fall detection method based on a millimeter-wave radar, which includes the steps of clustering point track data through a clustering algorithm to determine whether there is a target, and extracting relevant information of the target when there is a target; the point track data is the point track data in the geodetic coordinate system, which is converted from the point track data of the target output by the radar; the clustering steps are: (i) performing pre-clustering using velocity, calculating the absolute value of the difference between the velocities Vdi and Vdj of any two points, and clustering the points with this value less than the threshold deltaV into one class; (ii) performing clustering using distance, calculating the distance between any two points in the class obtained in step (i), and clustering the points with a distance less than the threshold Dist into one class; (iii) calculating the number of points in the class obtained in step (ii), and if the number is greater than the threshold Num, it is considered that there is a target; otherwise, it is considered that there is no target. The detection method of the present invention comprehensively considers multiple judgment factors and significantly improves the detection accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of radar applications, and relates to an indoor personnel fall detection method based on a millimeter-wave radar. Background Art

[0002] As a large populous country, China has a very serious problem of population aging. According to statistics, the total number of elderly people in China increased from 200 million to 250 million from 2015 to 2020, among which 118 million are empty nesters. According to the "Technical Guidelines for the Intervention of Elderly Falls" issued by the Ministry of Health, falls are the main cause of injury deaths among the elderly over 65 years old. The elderly have weak bodies. Once they fall, they may not be able to get up and ask for help by themselves. If the situation is serious and they cannot seek medical treatment in time, it will cause great harm to the elderly's bodies.

[0003] At present, some patent applications have disclosed personnel fall detection technologies based on millimeter-wave radars, but these detection technologies are not accurate.

[0004] Patent 202110572750.6 proposed "An Intelligent Detection System and Recognition Method for Human Falls Based on a Millimeter-Wave Radar", which combines the technologies of millimeter-wave radar detection + neural network. There are two defects in this application: (1) The scenarios of human falls are complex and changeable. For example, the layouts of bathrooms, offices, living rooms, etc. are different, and both stationary objects and moving objects will have reflection points, which leads to stationary objects interfering with moving objects. Each room of a user corresponds to a new scenario, resulting in a high false alarm rate when this method is applied to different scenarios; (2) When there are multiple moving targets in the room and the movement trajectories of the targets cross, it causes interference between multiple moving targets, resulting in a high false alarm rate.

[0005] Patent 202110233701.X and Patent 202011146676.3 both disclose fall detection methods based on millimeter-wave radars. The former mentions a device for detecting the fall of a target using a millimeter-wave radar, but does not mention how to implement the logic or algorithm therein; the latter mentions detecting the target posture according to the millimeter-wave radar to determine whether the target has fallen. After careful examination, it actually only uses the height information of the target, and the result is inaccurate.

[0006] In addition, when using a millimeter-wave radar for indoor personnel fall detection, since the dot traces reflected by the millimeter-wave radar are relatively sparse, the conclusions drawn based on the sparse dot traces are not reliable, and the disclosed technologies do not take this into account. Summary of the Invention

[0007] To solve the problems existing in the prior art, the present invention provides an indoor personnel fall detection method based on a millimeter-wave radar.

[0008] To achieve the above object, the solution adopted by the present invention is as follows:

[0009] An indoor personnel fall detection method based on millimeter-wave radar, comprising the steps of clustering point track data through a clustering algorithm to judge whether there is a target, and extracting relevant information of the target when there is a target;

[0010] The point track data is the point track data in the geodetic coordinate system, which is converted from the point track data of the target output by the radar;

[0011] The point track data of the target output by the radar includes the distance Range from the point to the radar, the point track velocity Vd, the point track horizontal angle α, and the point track pitch angle θ. The radar is a millimeter-wave radar, and the target is an indoor person;

[0012] The clustering steps are as follows:

[0013] (i) Use velocity pre-clustering, calculate the absolute value of the difference between the velocities Vdi and Vdj of any two points, and cluster the points with the absolute value of the velocity difference less than the threshold deltaV into one class. The units of Vdi and Vdj are m / s, and the value range of deltaV is 0.5 - 0.8 m / s;

[0014] (ii) Use distance clustering, calculate the distance between any two points in the class obtained in step (i), and cluster the points with the distance less than the threshold Dist into one class. The unit of distance is m, and the value range of Dist is 0.2 - 0.36 m;

[0015] (iii) Calculate the number of points in the class obtained in step (ii). If the number is greater than the threshold Num, it is considered that there is a target, and the value range of Num is 4 - 6; otherwise, it is considered that there is no target.

[0016] The present invention first uses velocity pre-clustering and then distance clustering. The reason for this design is that when using a millimeter-wave radar to detect indoor targets, in addition to people, there are many items in the indoor scene, such as sofas, TVs, tables and chairs, etc. These items will generate echoes, which are reflected in the detection of a large number of stationary point tracks by the radar. When a person moves next to these items, velocity pre-clustering can distinguish the point tracks reflected by the person from the point tracks reflected by the items. In addition, when the movement trajectories of indoor people cross and the velocity difference is greater than the threshold delta_V, velocity pre-clustering can also distinguish the point tracks reflected by different people. The order cannot be reversed because if distance pre-clustering is used first, the clustering result will contain a large number of point tracks emitted by non-people, increasing the computational complexity of clustering.

[0017] When performing velocity pre-clustering, the threshold delta_V is designed to be 0.5 to 0.8 m / s, which is a relatively small value. This is mainly because during the movement of the same person, the velocities of different echoes reflected to the radar module do not differ much. When performing distance clustering, the threshold Dist is designed to be 0.2 to 0.36 m, also considering that the distances of different echoes reflected to the radar module by the same person do not differ much.

[0018] As a preferred technical solution:

[0019] For an indoor personnel fall detection method based on millimeter-wave radar as described above, both velocity pre-clustering and distance clustering use the DBSCAN algorithm. The process of the DBSCAN algorithm is as follows:

[0020] (a) Calculate the distance d(i,j) between any two points in the sample set. When using velocity pre-clustering, the definition of the distance d(i,j) is as follows:

[0021] d(i,j) = |Vdi - Vdj|;

[0022] When using distance clustering, the definition of the distance d(i,j) is as follows:

[0023] d(i,j) = ((Xi - Xj)^2 + (Yi - Yj)^2 + (Zi - Zj)^2)^0.5;

[0024] In the formula, Xi, Yi, and Zi are the X-axis coordinate, Y-axis coordinate, and Z-axis coordinate of point i, respectively, with the unit of m; Xj, Yj, and Zj are the X-axis coordinate, Y-axis coordinate, and Z-axis coordinate of point j, respectively, with the unit of m.

[0025] (b) Count the number of points within the epsilon neighborhood of each point. When using velocity pre-clustering, the epsilon value is set to the threshold deltaV, and when using distance clustering, the epsilon value is set to the threshold Dist. If the number of points is greater than MinPoints, it means that this point is a core object, otherwise it is a clutter point. The value range of MinPoints is 3 to 5.

[0026] (c) Determine the density direct reach relationship. If point 1 is a core object and point 2 is within the epsilon neighborhood of point 1, then point 2 has a density direct reach to point 1.

[0027] (d) Determine the density reachable relationship. For any two points P and Q, if there exists a sequence of points X1, X2,......Xm that simultaneously satisfies: X1 is point P, Xm is point Q, Xt+1 has a density direct reach to Xt, m is an integer greater than 2, and 1 ≤ t ≤ m - 1, then point Q has a density reach to point P.

[0028] (e) Determine the density connectivity relationship. Point P and point Q are any two points. If there exists a core object point Xn such that point P and point Q are density-reachable, then point Q is density-connected to point P.

[0029] (f) Cluster all density-connected points into one cluster.

[0030] The DBSCAN algorithm is a typical density clustering method. Compared with other clustering methods, the DBSCAN algorithm has the following advantages: First, it does not require the number of classes to be set in advance; second, it can detect abnormal points while clustering, and the clustering algorithm is insensitive to abnormal points in the data set (the insensitivity to clutter points indicates that clutter will not have a significant impact on this clustering method).

[0031] In the present invention, the number of indoor targets is unknown in advance, meaning the number of classes cannot be pre-set, which perfectly complements the first advantage of DBSCAN. To address the issue of sparse point traces in millimeter-wave radar detection, the present invention combines N frames of data in preprocessing and outputs them (see below for details). Considering the phenomenon that, in a given N frames of data output, different frames may detect points emitted from different parts of the same person, with significant differences in distance between these points. In this case, DBSCAN's "density-reachable" search principle can effectively cluster these point traces into a single class, which is consistent with actual conditions.

[0032] In the above-mentioned method for detecting indoor falls of personnel based on millimeter-wave radar, the point trace data in the geodetic coordinate system are further merged before clustering.

[0033] The above-mentioned method for detecting indoor falls based on millimeter-wave radar includes the following steps:

[0034] (1) Design a queue container that can store N frames of data. The queue container is named BUFFER. When the radar starts scanning, it stores N consecutive frames of data in BUFFER. The value of N ranges from 3 to 4. The trace data exists in the form of a single frame. N-frame data represents the data that is merged together from N frames of trace data.

[0035] (2) When the number of frames scanned by the radar is updated, a new frame of data is stored in the BUFFER, and the earliest frame stored in the BUFFER is removed from the BUFFER;

[0036] (3) Output the data in BUFFER in real time.

[0037] As described above, the method for detecting indoor falls of a person based on millimeter-wave radar has the following steps:

[0038] S1: Initialize the radar system;

[0039] S2: The radar outputs the point track data of the target;

[0040] S3: Convert the point track data of the target output by the radar into point track data in the geodetic coordinate system;

[0041] S4: Merge the point track data in the geodetic coordinate system;

[0042] S5: Use the clustering algorithm to cluster the point track data to determine whether there is a target, and extract the relevant information of the target when there is a target;

[0043] S6: Judge the motion state of the target, and trigger the alarm module once it is detected that the motion state of the target is a fall.

[0044] For an indoor personnel fall detection method based on millimeter-wave radar as described above, in S3, the definition of the geodetic coordinate system is: the vertical projection of the radar installation position on the ground is used as the origin of the geodetic coordinate system, the vertical projection of the radar normal direction on the ground is used as the X-axis direction of the geodetic coordinate system, and the Y-axis and Z-axis directions in the geodetic coordinate system are determined according to the right-hand rule; the conversion uses the following equations:

[0045] Range_x = Range × (cosβcosθcosα + sinβsinθ);

[0046] Range_y = Range × cosθsinα;

[0047] Range_z = Range × (cosβsinθ - sinβcosθcosα) + h;

[0048] In the formula, Range_x, Range_y, and Range_z respectively represent the distance components of the target on the X-axis, Y-axis, and Z-axis in the geodetic coordinate system; Range represents the distance between the point and the radar, with the unit of m; β represents the depression angle (30°) between the radar normal and the horizontal plane during installation, with the unit of °; θ represents the pitch angle of the target, with a negative depression angle and a positive elevation angle, with the unit of °; α represents the horizontal angle of the target, with the unit of °; h is 2.2 - 2.4 m.

[0049] For an indoor personnel fall detection method based on millimeter-wave radar as described above, in S5, the relevant information of the target includes the instantaneous height H of the target, the instantaneous height difference delta_H, the average height ave_H, the average height difference delta_ave_H, the speed difference delta_Vd, and the acceleration A_Z. The units of H, delta_H, ave_H, and delta_ave_H are all m, the unit of delta_Vd is m / s, and the unit of A_Z is m / s2 ; The extraction steps are as follows:

[0050] (a) Divide the points belonging to the same class after clustering into M different regions according to the height value. The height value is the distance component of the target along the Z-axis in the geodetic coordinate system, and M = 4. Among them, region M1: height value > 1.85 m; region M2: 1.45 m < height value ≤ 1.85 m; region M3: 0.35 m < height value ≤ 1.45 m; region M4: height value ≤ 0.35 m. The basis for this classification is as follows: For the points in region M1, considering that this device is generally used to detect the falling state of the elderly living alone indoors, and the height of the elderly is generally less than 1.85 m, so the points in region M1 are considered as clutter points; for the classification of M2, M3, and M4, it is mainly considered that the speed differences at different heights during a fall are relatively large, and different weights should be multiplied when calculating the target speed in the subsequent steps.

[0051] (b) Extract the instantaneous height H of the target: Take the maximum and the second maximum values of the heights of the target in the class (i.e., the class obtained by clustering), and use their average value as the value of the instantaneous height H of the target.

[0052] (c) Extract the instantaneous height difference delta_H of the target: Subtract the instantaneous height H of the target at the previous moment from the instantaneous height H of the target at the current moment to obtain the instantaneous height difference delta_H of the target.

[0053] (d) Extract the average height ave_H of the target: Take the average value of the heights of the target in regions M2, M3, and M4 in step (a) as the average height ave_H of the target.

[0054] (e) Extract the average height difference delta_ave_H of the target: Subtract the average height ave_H of the target at the previous moment from the average height ave_H of the target at the current moment to obtain the average height difference delta_ave_H of the target.

[0055] (f) Extract the speed difference delta_Vd of the target: Perform a first-order difference on the speed Vd of the target to obtain the speed difference delta_Vd of the target. The calculation formula for the speed Vd of the target is as follows:

[0056]

[0057] In the formula, Vdk is the average value of the track speeds in the kth region, with the unit of m / s; Wk is the weight of the average value of the track speeds in the kth region, W1 = 0, W2 = 0.25, W3 = 0.5, W4 = 0.25;

[0058] (g) Extract the acceleration A_Z of the target, that is, the acceleration of the target in the Z direction: The calculation formula is as follows:

[0059] A_Z = (delta_ave_H1 – delta_ave_H2) / (delta_t)^2;

[0060] Wherein, delta_ave_H1 is the average height difference at the current moment, with the unit of m / s; delta_ave_H2 is the average height difference at the previous moment, with the unit of m / s; delta_t is the frame interval time of radar scanning, with the unit of s.

[0061] For an indoor personnel fall detection method based on millimeter-wave radar as described above, in S6, the process of judging the motion state of the target is as follows:

[0062] (1) Design a counter, named Counter, and initialize it to 0;

[0063] (2) Judge:

[0064] Condition 1: delta_H < 0;

[0065] Condition 2: delta_ave_H < 0;

[0066] Condition 3: delta_Vd > 0;

[0067] If at least two or more conditions are met, go to the next step; otherwise, return to step (1) and wait for the arrival of the next set of data;

[0068] When the radar detects the conditions in (2), it is considered that the human body starts to move downward in the Z-axis direction; if it is the process of falling, then the absolute value of the acceleration in the Z-axis direction will be relatively large; when this falling process is completed, the instantaneous height H and the average height ave_H of the human body will be relatively small; the above judgment actually uses the radar to judge the motion state of the human body from the start of falling to the end of falling;

[0069] In the above judgment process, the accuracy of the judgment is also improved as much as possible; in (2), theoretically, when the human body starts to move downward, the three judgment conditions will be met simultaneously, but the radar cannot guarantee that it detects the same part of the human body every time, so in the present invention, it is considered that when two or more of the three judgment conditions occur, it is considered that the human body starts to move downward in the Z-axis direction;

[0070] (3) Increment Counter by 1, record the two values with the largest and the second largest absolute values of A_Z, denoted as the maximum value A_Z1 and the second largest value A_Z2 (that is, the maximum value other than the absolute value maximum); theoretically, storing only one maximum value of the absolute value of acceleration is enough, but considering that the human body fall is a continuous process in the present invention, recording one maximum value and one second largest value also strengthens the determination of the fall process;

[0071] (4) Determine whether Counter is greater than the threshold value, where the threshold value ranges from 3 to 5. If so, proceed to the next step; otherwise, return to step (2) to continue the judgment for the next set of data.

[0072] (5) Determine whether both A_Z1 and A_Z2 are greater than the threshold value, where the threshold value ranges from 2 to 3 m / s 2 , if so, proceed to the next step (6); otherwise, return to step (2) to judge the next set of data;

[0073] (6) Judge:

[0074] Condition 1: The instantaneous height H of the target corresponding to the data in the real-time output BUFFER < the threshold value, where the threshold value ranges from 1.0 to 1.15 m;

[0075] Condition 2: The average height ave_H of the target corresponding to the data in the real-time output BUFFER < the threshold value, where the threshold value ranges from 0.5 to 0.8 m;

[0076] If both of the above two conditions are met simultaneously, output the motion state of the target as a fall; otherwise, output the motion state of the target as not fallen; the two judgment conditions are used to distinguish different situations such as human falls, sitting down, squatting down, lying on the sofa, etc.; after a human falls, both conditions will be met simultaneously, and other situations will not meet condition 2.

[0077] As described above, a method for indoor personnel fall detection based on millimeter-wave radar, the false alarm rate of the method for indoor personnel fall detection based on millimeter-wave radar is less than or equal to 20%, and the success rate is greater than or equal to 86%.

[0078] When using millimeter-wave radar for indoor personnel fall detection, the height information of the target is one of the most important features. However, the traces detected by the millimeter-wave radar are relatively sparse, and the output target height cannot reflect the true height of the target, resulting in inaccurate strategies formulated based on this most important feature; in addition, although the traces detected by the millimeter-wave radar are relatively sparse, when used for indoor target detection, due to the large number of indoor items and complex scenes, and the existence of false alarms and multipath effects in radar detection, the radar will detect a large amount of clutter information, and it becomes complex to identify the traces of personnel from so many clutter points. To solve the above problems, the present invention provides a new method for indoor personnel fall detection based on millimeter-wave radar, and its principle is as follows:

[0079] The present invention first merges data through a queue container, merges the N-frame track data continuously detected by the radar, and uses the N-frame track data to determine the instantaneous height and average height of the target; the tracks detected by the millimeter-wave radar are relatively sparse. By adopting the method of merging data in the present invention, as long as the true height of the target is detected in one of the N-frame track data, the output instantaneous height information of the target is accurate, which can significantly improve the accuracy of the target height information.

[0080] Secondly, aiming at the problem of a large number of clutters in the indoor scene, the present invention proposes a method of "velocity pre-clustering + distance clustering", which can identify the tracks reflected by indoor personnel from the environment with a large number of clutters; the echoes generated by indoor personnel during movement (walking, falling, etc.) have different speeds from the clutters. Using velocity pre-clustering, the tracks with similar speeds can be initially clustered into classes, so that static targets and moving targets can be distinguished. On this basis, using distance clustering, the tracks with similar positions can be clustered into classes; this clustering method can distinguish other indoor items from personnel, and can also distinguish personnel with different speeds.

[0081] Finally, when judging the falling state of the target, multiple features such as instantaneous height, instantaneous height difference, average height, average height difference, speed difference, and acceleration are comprehensively considered for judgment.

[0082] Compared with the disclosed technology, the present invention significantly improves the detection accuracy.

[0083] Beneficial effects

[0084] (1) A method for detecting indoor personnel falling based on millimeter-wave radar according to the present invention merges data through a queue container, solving the problem of sparse detection data of millimeter-wave radar.

[0085] (2) A method for detecting indoor personnel falling based on millimeter-wave radar according to the present invention creatively adopts a method of combining pre-clustering and clustering. First, the tracks that may belong to the same target are clustered into a class through velocity pre-clustering, and then through distance clustering, the performance of clustering is significantly improved, thereby improving the accuracy of the target information based on the clustered data.

[0086] (3) A method for detecting indoor personnel falling based on millimeter-wave radar according to the present invention comprehensively considers multiple judgment factors, and the performance is significantly improved compared with the currently disclosed technology. Description of the drawings

[0087] Figure 1 It is a flowchart of the method for detecting indoor personnel falling based on millimeter-wave radar according to the present invention.

[0088] Figure 2 It is a flowchart for judging the motion state of the target. Detailed implementation manners

[0089] The present invention will be further described below in conjunction with specific implementation manners. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by the appended claims of the present invention.

[0090] An indoor personnel fall detection method based on a millimeter-wave radar, as Figure 1 shown, the specific steps are as follows:

[0091] S1: Initialize the radar system; wherein, the radar is a millimeter-wave radar;

[0092] S2: The radar outputs the track data of the target; wherein, the track data includes the distance Range from the point to the radar, the track speed Vd, the track horizontal angle α, and the track pitch angle θ; the target is an indoor person;

[0093] S3: Convert the track data of the target output by the radar into track data in the geodetic coordinate system; wherein, the definition of the geodetic coordinate system is: taking the vertical projection of the radar installation position on the ground as the origin of the geodetic coordinate system, the vertical projection of the radar normal direction on the ground as the X-axis direction of the geodetic coordinate system, and the Y-axis and Z-axis directions in the geodetic coordinate system are determined according to the right-hand rule; the equations used for the conversion are as follows:

[0094] Range_x = Range × (cosβcosθcosα + sinβsinθ);

[0095] Range_y = Range × cosθsinα;

[0096] Range_z = Range × (cosβsinθ - sinβcosθcosα) + h;

[0097] In the formula, Range_x, Range_y, and Range_z respectively represent the distance components of the target on the X-axis, Y-axis, and Z-axis in the geodetic coordinate system; Range represents the distance from the point to the radar, with the unit of m; β represents the depression angle (30°) between the radar normal and the horizontal plane during installation, with the unit of °; θ represents the pitch angle of the target, with a negative depression angle and a positive elevation angle, with the unit of °; α represents the horizontal angle of the target, with the unit of °; h is 2.2 to 2.4 m;

[0098] S4: Merge the track data in the geodetic coordinate system. The specific merging steps are as follows:

[0099] (1) Design a queue container capable of storing N frames of data. This queue container is named BUFFER. When the radar starts scanning, consecutive N frames of data are sequentially stored in BUFFER, where the value range of N is 3 to 4;

[0100] (2) When the number of frames scanned by the radar is updated, a new frame of data is stored in BUFFER, and the earliest stored frame in BUFFER is removed from BUFFER;

[0101] (3) Output the data in BUFFER in real time;

[0102] S5: Use a clustering algorithm to cluster the track data to determine whether there is a target, and extract the relevant information of the target when there is a target; among them, the relevant information of the target includes the instantaneous height H of the target, the instantaneous height difference delta_H, the average height ave_H, the average height difference delta_ave_H, the speed difference delta_Vd, and the acceleration A_Z. The units of H, delta_H, ave_H, and delta_ave_H are all m, the unit of delta_Vd is m / s, and the unit of A_Z is m / s 2 ;

[0103] The clustering steps are as follows:

[0104] (i) Use velocity pre-clustering to calculate the absolute value of the difference between the velocities Vdi and Vdj of any two points, and cluster the points with the absolute value of the velocity difference less than the threshold deltaV into one class. The units of Vdi and Vdj are m / s, and the value range of deltaV is 0.5 to 0.8 m / s;

[0105] (ii) Use distance clustering to calculate the distance between any two points in the class obtained in step (i), and cluster the points with a distance less than the threshold Dist into one class. The unit of distance is m, and the value range of Dist is 0.2 to 0.36 m;

[0106] (iii) Calculate the number of points in the class obtained in step (ii). If the number is greater than the threshold Num, it is considered that there is a target, and the value range of Num is 4 to 6; otherwise, it is considered that there is no target;

[0107] Among them, both the velocity pre-clustering and the distance clustering use the DBSCAN algorithm. The process of the DBSCAN algorithm is as follows:

[0108] (a) Calculate the distance d(i,j) between any two points in the sample set. When using velocity pre-clustering, the definition of the distance d(i,j) is as follows:

[0109] d(i,j) = |Vdi - Vdj|;

[0110] When using distance clustering, the definition of the distance d(i,j) is as follows:

[0111] d(i,j) = ((Xi - Xj)^2+(Yi - Yj)^2+(Zi - Zj)^2)^0.5;

[0112] In the formula, Xi, Yi, and Zi are the X-axis coordinate, Y-axis coordinate, and Z-axis coordinate of point i respectively, and the unit of all is m; Xj, Yj, and Zj are the X-axis coordinate, Y-axis coordinate, and Z-axis coordinate of point j respectively, and the unit of all is m;

[0113] (b) Count the number of points within the epsilon neighborhood of each point. When using velocity pre-clustering, the epsilon value is set to the threshold deltaV, and when using distance clustering, the epsilon value is set to the threshold Dist. If the number of points is greater than MinPoints, it indicates that this point is a core object, otherwise it is a clutter point. The value range of MinPoints is 3 - 5;

[0114] (c) Determine the density direct reach relationship. If point 1 is a core object and point 2 is within the epsilon neighborhood of point 1, then point 2 density directly reaches point 1;

[0115] (d) Determine the density reachable relationship. For any two points P and Q, if there exists a sequence of points X1, X2,......Xm that simultaneously satisfies: X1 is point P, Xm is point Q, Xt+1 density directly reaches Xt, m is an integer greater than 2, and 1 ≤ t ≤ m - 1, then point Q density reaches point P;

[0116] (e) Determine the density connected relationship. For any two points P and Q, if there exists a core object point Xn such that point P and point Q density reach, then point Q is density connected to point P;

[0117] (f) Cluster all the density-connected points into one class;

[0118] The steps to extract the relevant information of the target are as follows:

[0119] (I) Divide the points belonging to the same class after clustering into M different regions according to the height value. The height value is the distance component of the Z-axis of the target in the geodetic coordinate system, and M = 4. Among them, region M1: height value > 1.85m; region M2: 1.45m < height value ≤ 1.85m; region M3: 0.35m < height value ≤ 1.45m; region M4: height value ≤ 0.35m;

[0120] (II) Extract the instantaneous height H of the target: Take the maximum value and the second maximum value of the height of the target in the class, and use their average value as the value of the instantaneous height H of the target;

[0121] (III) Extract the instantaneous height difference delta_H of the target: Perform a first-order difference on the instantaneous height H of the target obtained in step (II) to obtain the instantaneous height difference delta_H of the target;

[0122] (IV) Extract the average height ave_H of the target: Take the mean of the heights of the targets in regions M2, M3, and M4 in step (I) as the average height ave_H of the target;

[0123] (V) Extract the average height difference delta_ave_H of the target: Perform a first-order difference on the average height ave_H of the target obtained in step (IV) to obtain the average height difference delta_ave_H of the target;

[0124] (VI) Extract the velocity difference delta_Vd of the target: Perform a first-order difference on the velocity Vd of the target to obtain the velocity difference delta_Vd of the target, where the calculation formula for the velocity Vd of the target is as follows:

[0125]

[0126] In the formula, Vdk is the mean of the track velocities in the kth region, with the unit of m / s; Wk is the weight of the mean of the track velocities in the kth region, W1 = 0, W2 = 0.25, W3 = 0.5, W4 = 0.25;

[0127] (VII) Extract the acceleration A_Z of the target, that is, the acceleration of the target in the Z direction: The calculation formula is as follows:

[0128] A_Z = (delta_ave_H1 – delta_ave_H2) / (delta_t)^2;

[0129] In the formula, delta_ave_H1 is the average height difference at the current moment, with the unit of m / s; delta_ave_H2 is the average height difference at the previous moment, with the unit of m / s; delta_t is the frame interval time of radar scanning, with the unit of s;

[0130] S6: As Figure 2 shown, judge the motion state of the target, and trigger the alarm module once it is detected that the motion state of the target is a fall; among them, the process of judging the motion state of the target is as follows:

[0131] (i) Design a counter, named Counter, and initialize it to 0;

[0132] (ii) Judge:

[0133] Condition 1: delta_H < 0;

[0134] Condition 2: delta_ave_H < 0;

[0135] Condition 3: delta_Vd > 0;

[0136] If at least two or more conditions are met, proceed to the next step; otherwise, return to step (i) and wait for the arrival of the next set of data;

[0137] (iii) Increment Counter by 1, record the two values with the largest and second-largest absolute values of A_Z, denoted as the maximum value A_Z1 and the second-largest value A_Z2 (i.e., the maximum value other than the absolute value maximum);

[0138] (iv) Determine whether Counter is greater than the threshold, where the threshold ranges from 3 to 5. If so, proceed to the next step; otherwise, return to step (ii) and continue to judge the next set of data;

[0139] (v) Determine whether both A_Z1 and A_Z2 are greater than the threshold, where the threshold ranges from 2 to 3 m / s 2 , if so, proceed to the next step (vi); otherwise, return to step (ii) and judge the next set of data;

[0140] (vi) Judge:

[0141] Condition 1: The instantaneous height H of the target corresponding to the data in the real-time output BUFFER < the threshold, where the threshold ranges from 1.0 to 1.15 m;

[0142] Condition 2: The average height ave_H of the target corresponding to the data in the real-time output BUFFER < the threshold, where the threshold ranges from 0.5 to 0.8 m;

[0143] If the above two conditions are met simultaneously, output the motion state of the target as a fall; otherwise, output the motion state of the target as not fallen.

[0144] The false alarm rate of the indoor personnel fall detection method based on millimeter-wave radar is less than or equal to 20%, and the success rate is greater than or equal to 86%.

[0145] The following combines specific cases to illustrate the indoor personnel fall detection method based on millimeter-wave radar of the present invention. The present invention uses the Gatland Microelectronics RDP-77S244-ABM-AIP millimeter-wave radar sensor to test three different indoor scenarios: the bedroom, the living room, and the bathroom; in the method of the present invention, the value of N in S4 is 4, the value of deltaV in S5 is 0.8 m / s, the value of Dist is 0.36 m, the value of Num is 6, the value of MinPoints is 5, the threshold value in (iv) of S6 is 5, and the threshold value in (v) is 3 m / s2 For condition 1 in (vi), the threshold value is 1.15 m, and for condition 2 in (vi), the threshold value is 0.8 m.

[0146] Each scenario is tested 200 times, with 100 times for testing the false positive rate and the other 100 times for testing the success rate. The calculation methods for the false positive rate FPR and the success rate TRP are as follows:

[0147]

[0148]

[0149] Among them, T1 is the number of times the output status is a fall after performing actions (squatting 35 times, sitting 35 times, walking 30 times) in a certain scenario, and T2 is the number of times the output status is a fall after performing actions (falling 100 times) in a certain scenario. As shown in Table 1, the method disclosed in the present invention, compared with the disclosed methods (methods that do not consider merging radar output data, do not use velocity pre-clustering, and only consider target height information, such as the methods mentioned in Patent 202110233701.X and Patent 202011146676.3), significantly reduces the false positive rate and greatly improves the detection success rate.

[0150] Table 1

[0151]

Claims

1. An indoor personnel fall detection method based on millimeter-wave radar, characterized in that, The steps of the indoor personnel fall detection method based on millimeter-wave radar are as follows: S1: Initialize the radar system; S2: The radar outputs the track data of the target; The track data of the target output by the radar includes the distance Range from the point to the radar, the track velocity Vd, the track horizontal angle α, and the track pitch angle θ. The radar is a millimeter-wave radar, and the target is an indoor person; S4: Convert the track data of the target output by the radar into track data in the geodetic coordinate system; S5: Merge the track data in the geodetic coordinate system; S6: Use a clustering algorithm to cluster the track data to determine whether there is a target, and extract relevant information of the target when there is a target; The clustering steps are as follows: (i) Use velocity pre-clustering. Calculate the absolute value of the difference between the velocities Vdi and Vdj of any two points, and cluster the points with the absolute value of the velocity difference less than the threshold deltaV into one class. The units of Vdi and Vdj are m / s, and the value range of deltaV is 0.5 - 0.8 m / s; (ii) Use distance clustering. Calculate the distance between any two points in the class obtained in step (i), and cluster the points with the distance less than the threshold Dist into one class. The unit of the distance is m, and the value range of Dist is 0.2 - 0.36 m; (iii) Calculate the number of points in the class obtained in step (ii). If the number is greater than the threshold Num, it is considered that there is a target, and the value range of Num is 4 - 6; otherwise, it is considered that there is no target; The relevant information of the target includes the instantaneous height H of the target, the instantaneous height difference delta_H, the average height ave_H, the average height difference delta_ave_H, the velocity difference delta_Vd, and the acceleration A_Z. The units of H, delta_H, ave_H, and delta_ave_H are all m, the unit of delta_Vd is m / s, and the unit of A_Z is m / s 2 ; S6: Judge the motion state of the target. Once it is detected that the motion state of the target is a fall, trigger the alarm module. The process of judging the motion state of the target is as follows: (1) Design a counter named Counter and initialize it to 0; (2) Judge: Condition 1: delta_H < 0; Condition 2: delta_ave_H < 0; Condition 3: delta_Vd > 0; If at least two of the above conditions are met, go to the next step; otherwise, return to step (1); (3) Increment Counter by 1, record the two values with the largest and second-largest absolute values of A_Z, denoted as the maximum value A_Z1 and the second-largest value A_Z2; (4) Judge whether Counter is greater than the threshold. The value range of the threshold is 3 - 5. If so, go to the next step; otherwise, return to step (2); (5) Determine whether both A_Z1 and A_Z2 are greater than the threshold value, and the value range of the threshold is 2 to 3 m / s 2 , if so, proceed to the next step (6); otherwise, return to step (2); (6) Judge: Condition 1: The H corresponding to the data in the real-time output BUFFER < the threshold. The value range of the threshold is 1.0 - 1.15 m; Condition 2: The ave_H corresponding to the data in the real-time output BUFFER < the threshold. The value range of the threshold is 0.5 - 0.8 m; If both of the above conditions are met at the same time, output the motion state of the target as a fall; otherwise, output the motion state of the target as not fallen.

2. The indoor personnel fall detection method based on millimeter-wave radar according to claim 1, wherein, Both the velocity pre-clustering and the distance clustering use the DBSCAN algorithm. The process of the DBSCAN algorithm is as follows: (a) Calculate the distance d(i,j) between any two points in the sample set. The definition of the distance d(i,j) when using velocity pre-clustering is as follows: d(i,j) = |Vdi - Vdj|; When using distance clustering, the definition of the distance d(i,j) is as follows: d(i,j) = ((Xi - Xj)^2 + (Yi - Yj)^2 + (Zi - Zj)^2)^0.5; In the formula, Xi, Yi, and Zi are the X-axis coordinate, Y-axis coordinate, and Z-axis coordinate of point i respectively, and the unit of all is m; Xj, Yj, and Zj are the X-axis coordinate, Y-axis coordinate, and Z-axis coordinate of point j respectively, and the unit of all is m; (b) Count the number of points within the epsilon neighborhood of each point. When using velocity pre-clustering, the epsilon value is set to the threshold deltaV, and when using distance clustering, the epsilon value is set to the threshold Dist. If the number of points is greater than MinPoints, it indicates that this point is a core object, otherwise it is a clutter point. The value range of MinPoints is 3 - 5; (c) Determine the density direct reach relationship. If point 1 is a core object and point 2 is within the epsilon neighborhood of point 1, then point 2 has a density direct reach to point 1; (d) Determine the density reachable relationship. For any two points P and Q, if there exists a sequence of points X1, X2,......Xm that simultaneously satisfies: X1 is point P, Xm is point Q, Xt+1 has a density direct reach to Xt, m is an integer greater than 2, and 1 ≤ t ≤ m - 1, then point Q has a density reach to point P; (e) Determine the density connected relationship. For any two points P and Q, if there exists a core object point Xn such that points P and Q are density reachable, then points Q and P are density connected; (f) Cluster all density-connected points into one class.

3. A method for indoor personnel fall detection based on millimeter-wave radar according to claim 1, characterized in that Before clustering, the point trace data in the geodetic coordinate system is also merged.

4. A method for indoor personnel fall detection based on millimeter-wave radar according to claim 3, characterized in that The merging steps are as follows: (1) Design a queue container that can store N frames of data. This queue container is named BUFFER. When the radar starts scanning, consecutive N frames of data are sequentially stored in BUFFER. The value range of N is 3 - 4; (2) When the number of frames scanned by the radar is updated, a new frame of data is stored in BUFFER, and the earliest stored frame in BUFFER is removed from BUFFER; (3) Output the data in BUFFER in real time.

5. A method for indoor personnel fall detection based on millimeter-wave radar according to claim 1, characterized in that, In S3, the definition of the geodetic coordinate system is: Taking the vertical projection of the radar installation position on the ground as the origin of the geodetic coordinate system, and the vertical projection of the radar normal direction on the ground as the X-axis direction of the geodetic coordinate system. The Y-axis and Z-axis directions in the geodetic coordinate system are determined according to the right-hand rule; The equations used for the transformation are as follows: Range_x = Range × (cosβcosθcosα + sinβsinθ); Range_y = Range × cosθsinα; Range_z = Range × (cosβsinθ - sinβcosθcosα) + h; In the formula, Range_x, Range_y, and Range_z respectively represent the distance components of the target on the X-axis, Y-axis, and Z-axis in the geodetic coordinate system; Range represents the distance from the point to the radar, and the unit is m; β represents the depression angle between the radar normal and the horizontal plane during installation, and the unit is °; h is 2.2 - 2.4m.

6. The indoor personnel fall detection method based on millimeter wave radar according to claim 1, characterized in that, In S5, the extraction steps are as follows: (a) Divide the points belonging to the same class after clustering into M different regions according to the height value. The height value is the distance component of the target along the Z-axis in the geodetic coordinate system, and M = 4. Among them, region M1: height value > 1.85 m; region M2: 1.45 m < height value ≤ 1.85 m; region M3: 0.35 m < height value ≤ 1.45 m; region M4: height value ≤ 0.35 m; (b) Extract the instantaneous height H of the target: Take the maximum and the second maximum values of the heights of the targets in the class, and use their average value as the value of H; (c) Extract the instantaneous height difference delta_H of the target: Subtract the H of the previous moment from the H of the current moment to get delta_H; (d) Extract the average height ave_H of the target: Take the average value of the heights of the targets in regions M2, M3, and M4 in step (a) as ave_H; (e) Extract the average height difference delta_ave_H of the target: Subtract the ave_H of the previous moment from the ave_H of the current moment to get delta_ave_H; (f) Extract the velocity difference delta_Vd of the target: Perform a first-order difference on the velocity Vd of the target to get delta_Vd, where the calculation formula of Vd is as follows: In the formula, Vdk is the average value of the track velocities in the k-th region, with the unit of m / s; Wk is the weight of the average value of the track velocities in the k-th region, W1 = 0, W2 = 0.25, W3 = 0.5, W4 = 0.25; (g) Extract the acceleration A_Z of the target: The calculation formula is as follows: A_Z = (delta_ave_H1 – delta_ave_H2) / (delta_t)^2; In the formula, delta_ave_H1 is the average height difference at the current moment, with the unit of m / s; delta_ave_H2 is the average height difference at the previous moment, with the unit of m / s; delta_t is the frame interval time of radar scanning, with the unit of s.

7. A method for indoor personnel fall detection based on millimeter-wave radar according to claim 1, characterized in that, The false alarm rate of the indoor personnel fall detection method based on millimeter-wave radar is less than or equal to 20%, and the success rate is greater than or equal to 86%.

Images