Fall detection method, device and electronic device

Abstract

The present invention provides a fall detection method, device and electronic device. The method comprises: obtaining point cloud features of a target to be detected in a point cloud set of each of a first preset number of continuous radar frames, and obtaining a first point cloud feature sequence that sequentially records the point cloud features of each of the first preset number of radar frames; inputting the first point cloud feature sequence into a preset recurrent neural network model, and obtaining a behavior category probability array of the first point cloud feature sequence; wherein the preset recurrent neural network model is trained based on point cloud feature sequence samples of multiple behavior categories that are pre-labeled, and the behavior category probability array includes probability values ​​of each behavior category; generating a fall detection result according to the behavior category probability array of the first point cloud feature sequence. The fall detection method provided by the present invention has the advantages of being easy to use and having a high detection accuracy.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a fall detection method, device and electronic equipment. Background Art

[0002] As the aging phenomenon continues to intensify, the safety of elderly people living alone has gradually become one of the focuses of society. Elderly people living alone may encounter a variety of safety accidents in their daily lives, among which safety accidents caused by falls account for a large proportion.

[0003] At present, contact devices and non-contact devices are usually used to detect falling behavior. Contact devices can be devices such as accelerometers, which can determine whether a person's center of gravity has an accelerated downward trend within a set time, thereby detecting whether a person has fallen. Non-contact devices can be devices such as millimeter wave radars, which can collect a person's height information within a set time, and then calculate the height difference and height parameters such as the speed of height change and acceleration, thereby detecting whether a person has fallen.

[0004] However, contact devices need to be carried around by people, which is inconvenient to use. For non-contact devices, due to falls by different people, different positions, and even different postures, the calculated height parameters are not stable, which is prone to misjudgment. In addition, for squatting, sitting, bending over, and other behaviors that are extremely similar to the height changes in the fall process, it is difficult to distinguish, which is also prone to misjudgment. Therefore, there is an urgent need for a fall detection method that is easy to use and has a high detection accuracy. Summary of the invention

[0005] Embodiments of the present invention provide a fall detection method, device and electronic device to solve the problem of lack of a fall detection method with high usability and accuracy.

[0006] In a first aspect, an embodiment of the present invention provides a fall detection method, comprising:

[0007] Acquire point cloud features of the target to be detected in point cloud sets of each of the first preset number of continuous radar frames, and obtain a first point cloud feature sequence that sequentially records the point cloud features of each of the first preset number of radar frames;

[0008] Inputting the first point cloud feature sequence into a preset recurrent neural network model to obtain a behavior category probability array of the first point cloud feature sequence; wherein the preset recurrent neural network model is trained based on point cloud feature sequence samples of multiple behavior categories that are pre-labeled, the behavior categories include falling behavior, and at least one of squatting behavior, sitting behavior, bending behavior, climbing behavior, or walking straight behavior, and the behavior category probability array includes probability values ​​of each behavior category;

[0009] Generate a fall detection result according to the behavior category probability array of the first point cloud feature sequence.

[0010] In a possible implementation, before acquiring point cloud features of the target to be detected in the point cloud sets of each of the first preset number of consecutive radar frames, the fall detection method further includes:

[0011] Acquiring a first preset number of continuous radar frames;

[0012] The point cloud set of a single acquired radar frame is determined as the point cloud set of the corresponding radar frame; or, the point cloud set of the acquired target radar frame and the point cloud set of the last radar frame acquired before the target radar frame are mapped to the same coordinate system, and the mapped point cloud set is determined as the point cloud set of the target radar frame; wherein the target radar frame is any one of the first preset number of radar frames.

[0013] In a possible implementation, the point cloud features include point count features, size features, and centroid features of the target to be detected; obtaining the point cloud features of the target to be detected in the point cloud sets of each of the first preset number of consecutive radar frames includes:

[0014] Determine all points corresponding to the target to be detected from the point cloud set;

[0015] The number of all points is determined as the point number feature of the target to be detected;

[0016] Obtain the maximum X-axis coordinate, the maximum Y-axis coordinate, the maximum Z-axis coordinate, the minimum X-axis coordinate, the minimum Y-axis coordinate, and the minimum Z-axis coordinate of all the point coordinates, and determine the difference between the maximum X-axis coordinate and the minimum X-axis coordinate, the difference between the maximum Y-axis coordinate and the minimum Y-axis coordinate, and the difference between the maximum Z-axis coordinate and the minimum Z-axis coordinate as the size feature of the target to be detected;

[0017] Accumulate the first product value of the X coordinate of each point and the signal-to-noise ratio, the second product value of the Y coordinate of each point and the signal-to-noise ratio, and the third product value of the Z coordinate of each point and the signal-to-noise ratio respectively, and calculate the first quotient value of the sum of the accumulated first product values ​​and the number of all points, the second quotient value of the sum of the second product values ​​and the number of all points, and the third quotient value of the sum of the third product values ​​and the number of all points respectively, and determine the first quotient value, the second quotient value, and the third quotient value as the centroid feature of the target to be detected.

[0018] In a possible implementation, before inputting the first point cloud feature sequence into the preset recurrent neural network model, the fall detection method further includes:

[0019] Normalizing the first point cloud feature sequence to obtain a normalized first point cloud feature sequence;

[0020] Input the first point cloud feature sequence into a preset recurrent neural network model, including:

[0021] The normalized first point cloud feature sequence is input into the preset recurrent neural network model.

[0022] In a possible implementation, generating a fall detection result according to the behavior category probability array of the first point cloud feature sequence includes:

[0023] Obtain the behavior category corresponding to the maximum probability value in the behavior category probability array of the first point cloud feature sequence;

[0024] If the behavior category corresponding to the maximum probability value is a fall behavior, a fall detection result indicating that a fall exists is generated; otherwise, a fall detection result indicating that a fall does not exist is generated.

[0025] In a possible implementation, generating a fall detection result according to the behavior category probability array of the first point cloud feature sequence includes:

[0026] generating a fall detection result according to the behavior category probability array of the first point cloud feature sequence and the behavior category probability array of at least one second point cloud feature sequence acquired in advance;

[0027] Among them, at least one second point cloud feature sequence is an adjacent point cloud feature sequence acquired before the first point cloud feature sequence, the adjacent point cloud feature sequence sequentially records the point cloud features of each of a first preset number of consecutive radar frames, and the adjacent point cloud feature sequence and the point cloud feature sequence first acquired thereafter have the following characteristics: all radar frames corresponding to the second point cloud feature to the last point cloud feature recorded sequentially in the adjacent point cloud feature sequence are the same as all radar frames corresponding to the first point cloud feature to the second to last point cloud feature recorded sequentially in the point cloud feature sequence first acquired thereafter.

[0028] In a possible implementation, generating a fall detection result according to the behavior category probability array of the first point cloud feature sequence and the behavior category probability array of at least one second point cloud feature sequence acquired in advance includes:

[0029] Determine the behavior category corresponding to the maximum probability value in the behavior category probability array of the target point cloud feature sequence as the behavior category of the target point cloud feature sequence; wherein the target point cloud feature sequence is any one of the first point cloud feature sequence and at least one second point cloud feature sequence;

[0030] If the proportion of the target point cloud feature sequences whose behavior category is fall behavior to all target point cloud feature sequences exceeds a preset threshold, a fall detection result indicating the presence of a fall is generated; otherwise, a fall detection result indicating the absence of a fall is generated.

[0031] In a possible implementation, after generating the fall detection result, the fall detection method further includes:

[0032] Acquire a height value of a radar frame corresponding to the last point cloud feature recorded in sequence in the first point cloud feature sequence, and a height value of each of a second preset number of radar frames acquired successively before the radar frame corresponding to the last point cloud feature;

[0033] Determine, among all the altitude values, a first radar frame where the maximum altitude value is located and a second radar frame where the minimum altitude value is located;

[0034] If the maximum height value is greater than the first preset height, the minimum height value is less than the second preset height, and the time difference between the first radar frame and the second radar frame is greater than the preset time difference, a fall alarm message is output.

[0035] In a second aspect, an embodiment of the present invention provides a fall detection device, including:

[0036] An acquisition module is used to acquire point cloud features of the target to be detected in the point cloud sets of each of the first preset number of radar frames, and obtain a first point cloud feature sequence that sequentially records the point cloud features of each of the first preset number of radar frames; wherein the point cloud features include point count features, size features, and centroid features of the target to be detected;

[0037] An input module is used to input the first point cloud feature sequence into a preset recurrent neural network model to obtain a behavior category probability array of the first point cloud feature sequence; wherein the preset recurrent neural network model is trained based on point cloud feature sequence samples of multiple behavior categories that are pre-labeled, and the behavior categories include falling behavior, and at least one of squatting behavior, sitting behavior, bending behavior, climbing behavior, or walking straight behavior, and the behavior category probability array includes probability values ​​of each behavior category;

[0038] The detection module is used to generate a fall detection result according to the behavior category probability array of the first point cloud feature sequence.

[0039] In a possible implementation, the acquisition module is further used to:

[0040] Acquiring a first preset number of continuous radar frames;

[0041] The point cloud set of a single acquired radar frame is determined as the point cloud set of the corresponding radar frame; or, the point cloud set of the acquired target radar frame and the point cloud set of the last radar frame acquired before the target radar frame are mapped to the same coordinate system, and the mapped point cloud set is determined as the point cloud set of the target radar frame; wherein the target radar frame is any one of the first preset number of radar frames.

[0042] In a possible implementation, the acquisition module is further used to:

[0043] Determine all points corresponding to the target to be detected from the point cloud set;

[0044] The number of all points is determined as the point number feature of the target to be detected;

[0045] Obtain the maximum X-axis coordinate, the maximum Y-axis coordinate, the maximum Z-axis coordinate, the minimum X-axis coordinate, the minimum Y-axis coordinate, and the minimum Z-axis coordinate of all the point coordinates, and determine the difference between the maximum X-axis coordinate and the minimum X-axis coordinate, the difference between the maximum Y-axis coordinate and the minimum Y-axis coordinate, and the difference between the maximum Z-axis coordinate and the minimum Z-axis coordinate as the size feature of the target to be detected;

[0046] Accumulate the first product value of the X coordinate of each point and the signal-to-noise ratio, the second product value of the Y coordinate of each point and the signal-to-noise ratio, and the third product value of the Z coordinate of each point and the signal-to-noise ratio respectively, and calculate the first quotient value of the sum of the accumulated first product values ​​and the number of all points, the second quotient value of the sum of the second product values ​​and the number of all points, and the third quotient value of the sum of the third product values ​​and the number of all points respectively, and determine the first quotient value, the second quotient value, and the third quotient value as the centroid feature of the target to be detected.

[0047] In a possible implementation, the fall detection system further includes a normalization module, which is used to:

[0048] Normalizing the first point cloud feature sequence to obtain a normalized first point cloud feature sequence;

[0049] Input the first point cloud feature sequence into a preset recurrent neural network model, including:

[0050] The normalized first point cloud feature sequence is input into the preset recurrent neural network model.

[0051] In a possible implementation, the detection module is further used to:

[0052] Obtain the behavior category corresponding to the maximum probability value in the behavior category probability array of the first point cloud feature sequence;

[0053] If the behavior category corresponding to the maximum probability value is a fall behavior, a fall detection result indicating that a fall exists is generated; otherwise, a fall detection result indicating that a fall does not exist is generated.

[0054] In a possible implementation, the detection module is further used to:

[0055] generating a fall detection result according to the behavior category probability array of the first point cloud feature sequence and the behavior category probability array of at least one second point cloud feature sequence acquired in advance;

[0056] Among them, at least one second point cloud feature sequence is an adjacent point cloud feature sequence acquired before the first point cloud feature sequence, the adjacent point cloud feature sequence sequentially records the point cloud features of each of a first preset number of consecutive radar frames, and the adjacent point cloud feature sequence and the point cloud feature sequence first acquired thereafter have the following characteristics: all radar frames corresponding to the second point cloud feature to the last point cloud feature recorded sequentially in the adjacent point cloud feature sequence are the same as all radar frames corresponding to the first point cloud feature to the second to last point cloud feature recorded sequentially in the point cloud feature sequence first acquired thereafter.

[0057] In a possible implementation, the detection module is further used to:

[0058] Determine the behavior category corresponding to the maximum probability value in the behavior category probability array of the target point cloud feature sequence as the behavior category of the target point cloud feature sequence; wherein the target point cloud feature sequence is any one of the first point cloud feature sequence and at least one second point cloud feature sequence;

[0059] If the proportion of the target point cloud feature sequences whose behavior category is fall behavior to all target point cloud feature sequences exceeds a preset threshold, a fall detection result indicating the presence of a fall is generated; otherwise, a fall detection result indicating the absence of a fall is generated.

[0060] In a possible implementation, the fall detection system further includes an output module for:

[0061] Acquire a height value of a radar frame corresponding to the last point cloud feature recorded in sequence in the first point cloud feature sequence, and a height value of each of a second preset number of radar frames acquired successively before the radar frame corresponding to the last point cloud feature;

[0062] Determine, among all the altitude values, a first radar frame where the maximum altitude value is located and a second radar frame where the minimum altitude value is located;

[0063] If the maximum height value is greater than the first preset height, the minimum height value is less than the second preset height, and the time difference between the first radar frame and the second radar frame is greater than the preset time difference, a fall alarm message is output.

[0064] In a third aspect, an embodiment of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method described in the first aspect when executing the computer program.

[0065] The embodiment of the present invention provides a fall detection method, device and electronic device, firstly obtaining the point cloud features of the target to be detected in the point cloud set of each of the first preset number of radar frames, and obtaining a first point cloud feature sequence which sequentially records the point cloud features of each of the first preset number of radar frames. Then, the first point cloud feature sequence is input into a preset recurrent neural network model to obtain a behavior category probability array of the first point cloud feature sequence. Finally, a fall detection result is generated according to the behavior category probability array of the first point cloud feature sequence.

[0066] Since the point cloud features can include the point count features, size features and centroid features of the target to be detected, and cover the height change, size and other information of the human body to be detected, the point cloud features can more comprehensively describe the instantaneous state and process of the human body falling. In addition, the preset recurrent neural network model adopted is trained based on point cloud feature sequence samples of various behavior categories that are pre-labeled. It can well distinguish interfering movements such as squatting, sitting, and bending over from falls, so the accuracy of fall detection is high. In addition, the fall detection method provided in the embodiment of the present invention can be integrated on the millimeter wave radar, without the need for the user to carry it with him, and has the advantages of high ease of use. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0068] Figure 1 is a flowchart of a fall detection method provided by an embodiment of the present invention;

[0069] Figure 2 is a structural schematic diagram of a fall detection device provided by an embodiment of the present invention;

[0070] Figure 3 It is a schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0071] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0072] In order to make the purpose, technical solutions and advantages of the present invention more clear, specific embodiments will be described below in conjunction with the accompanying drawings.

[0073] As described in the background technology, currently, contact devices and non-contact devices are usually used to detect fall behavior. However, both contact devices and non-contact devices have corresponding problems, and a fall detection method that is easy to use and has a high detection accuracy is urgently needed.

[0074] In order to solve the problems in the prior art, the embodiments of the present invention provide a fall detection method, device and electronic device. The fall detection method provided by the embodiments of the present invention is first introduced below.

[0075] The executor of the fall detection method may be a fall detection device, which may be an electronic device having a processor and a memory, such as a millimeter wave radar, and is not specifically limited in the embodiment of the present invention.

[0076] See also Figure 1 , which shows a flow chart of implementing the fall detection method provided by an embodiment of the present invention, comprising the following steps:

[0077] Step S110: acquiring point cloud features of the target to be detected in the point cloud sets of each of the first preset number of continuous radar frames, and obtaining a first point cloud feature sequence that sequentially records the point cloud features of each of the first preset number of radar frames.

[0078] In the field of electromagnetism, electronic devices such as transmitters can emit electromagnetic waves, which are refracted and reflected multiple times in the natural environment and then received by electronic devices such as receivers. Through signal processing methods, the information carried by electromagnetic waves can be parsed, such as the distance and angle from targets such as people, animals, and vehicles in the environment to the transmitter. When performing signal processing, radar frames are usually used as processing units. Each radar frame includes a large number of points that can reflect information such as target distance and speed. The sum of these points can be called a point cloud set.

[0079] In some embodiments, the first preset number may be any preset number, such as 10 or 20. The first preset number of continuous radar frames may be a corresponding number of radar frames continuously generated when fall detection is performed on a detection object such as an elderly person.

[0080] Considering that the point cloud in the radar frame corresponding to the millimeter-wave radar is usually sparse, it is difficult for the point cloud of a single radar frame to describe the target, which is not conducive to subsequent fall detection. In this regard, a neighboring frame accumulation scheme can be used to increase the point cloud in the radar frame.

[0081] Specifically, a first preset number of continuous radar frames are obtained, and then the point cloud set of the obtained target radar frame and the point cloud set of the last radar frame obtained before the target radar frame are mapped to the same coordinate system, and the mapped point cloud set is determined as the point cloud set of the target radar frame; wherein the target radar frame is any radar frame in the first preset number of radar frames. Since the human body moves slowly, motion compensation processing is not required.

[0082] The following takes the i-th radar frame as an example to refer to the target radar frame, and gives a point cloud set of the target radar frame. The point cloud set can be expressed as pointsInfo, as follows:

[0083] pointsInfo={(x(1),y(1),z(1)),(x(2),y(2),z(2)),……,

[0084] (x(totalNum_0),y(totalNum_0),z(totalNum_0)),

[0085] (x(totalNum_0+1),y(totalNum_0+1),z(totalNum_0+1)),……

[0086] (x(totalNum_0+totalNum_1),y(totalNum_0+totalNum_1),z(totalNum_0+totalNum_1))}

[0087] Among them, totalNum_0 represents the total number of points in the point cloud set in the i-1th radar frame, totalNum_1 represents the total number of points in the point cloud set in the i-th radar frame, and totalNum represents the total number of points obtained after mapping the i-1th radar frame and the i-th radar frame. So we can get:

[0088] totalNum=totalNum_0+totalNum_1.

[0089] In some embodiments, in order to improve data processing efficiency, the point cloud set of a single radar frame may also be directly determined as the point cloud set of the corresponding radar frame.

[0090] In some embodiments, after obtaining the point cloud sets of the first preset number of consecutive radar frames, clustering and track processing can be performed on the point cloud sets of each radar frame to extract the point cloud features of the target to be detected in the corresponding point cloud set, such as the point count feature, size feature, and centroid feature of the target to be detected. Afterwards, the point cloud features of the first preset number of radar frames can be recorded in sequence into a feature sequence, which can be referred to as a first point cloud feature sequence.

[0091] It should be noted that since the points in a point cloud set are usually disordered and difficult to use directly, feature extraction can be performed on the point cloud set. The extracted point cloud features can not only characterize the overall shape or relative relationship of the point cloud, but also facilitate subsequent regularization and dimensionality reduction processing.

[0092] The following is a detailed introduction to the point cloud feature extraction process. Taking the point cloud set of a radar frame as an example, first, all points corresponding to the target to be detected are determined from the point cloud set. After that, the number of all the above points is determined as the point number feature of the target to be detected. Next, the maximum X-axis coordinate, the maximum Y-axis coordinate, the maximum Z-axis coordinate, the minimum X-axis coordinate, the minimum Y-axis coordinate and the minimum Z-axis coordinate of the point coordinates of all the above points are obtained, and then the difference between the maximum X-axis coordinate and the minimum X-axis coordinate, the difference between the maximum Y-axis coordinate and the minimum Y-axis coordinate, and the difference between the maximum Z-axis coordinate and the minimum Z-axis coordinate are determined as the size feature of the target to be detected. Finally, the first product value of the X coordinate of each point and the signal-to-noise ratio, the second product value of the Y coordinate of each point and the signal-to-noise ratio, and the third product value of the Z coordinate of each point and the signal-to-noise ratio are accumulated respectively, and the first quotient value of the sum of the accumulated first product values ​​and the number of all points, the second quotient value of the sum of the second product values ​​and the number of all points, and the third quotient value of the sum of the third product values ​​and the number of all points are calculated respectively, and then the first quotient value, the second quotient value, and the third quotient value are determined as the centroid features of the target to be detected.

[0093] Specifically, the point number feature can be expressed as totalNum, the size feature can be expressed as xsize, ysize, zsize, and the center of mass feature can be expressed as xcenter, ycenter, zcenter. The corresponding calculation formula can be as follows:

[0094] xsize=max(X)–min(X);

[0095] ysize=max(Y)–min(Y);

[0096] zsize=max(Z)–min(Z);

[0097]

[0098]

[0099]

[0100] Among them, X represents all X-axis coordinates in the point cloud set, Y represents all Y-axis coordinates in the point cloud set, and Z represents all Z-axis coordinates in the point cloud set. i Represents the signal-to-noise ratio of the i-th point.

[0101] Optionally, in order to reduce the dimensional influence between different point cloud features, before performing the following step S120, the first point cloud feature sequence may be normalized.

[0102] Specifically, for the point feature, it can be normalized in the following way, namely:

[0103]

[0104] Among them, totalNum_normal represents the normalized point feature, and maxPoints represents the maximum number of points in the point cloud set.

[0105] For the size feature, it can be normalized in the following way, namely:

[0106]

[0107]

[0108]

[0109] Among them, xsize_normal, ysize_normal, and zsize_normal represent the normalized size features, and maxHeight represents the maximum Z coordinate in the point cloud set. It should be noted that the maximum Z coordinate is the highest point of the body of the target to be detected when it is moving straight.

[0110] For the centroid feature, it can be normalized in the following way, namely:

[0111]

[0112]

[0113]

[0114] Among them, xcenter_normal, ycenter_normal, and zcenter_normal represent the normalized centroid features, and maxRange is the distance from the farthest point to the radar in the detection area corresponding to the point cloud set. In this way, the interval (-maxRange, maxRange) can be normalized to the interval (0, 1).

[0115] In this way, after all the point number features, size features and centroid features in the first point cloud feature sequence are normalized, a normalized first point cloud feature sequence can be obtained.

[0116] Step S120: input the first point cloud feature sequence into a preset recurrent neural network model to obtain a behavior category probability array of the first point cloud feature sequence.

[0117] In some embodiments, the first point cloud feature sequence input to the preset recurrent neural network model may be the above-mentioned point cloud feature sequence after normalization, or may be a point cloud feature sequence that has not been normalized, which is not specifically limited here. It should be noted that the point cloud feature sequence after normalization can achieve a better detection effect, and the point cloud feature sequence that has not been normalized can achieve a higher data processing efficiency, and the corresponding type of point cloud feature sequence can be selected according to different needs.

[0118] In some embodiments, the preset recurrent neural network model can be a model trained based on point cloud feature sequence samples of multiple behavior categories that are pre-labeled. The model can classify the input first point cloud feature sequence and obtain the probability values ​​of the first point cloud feature sequence belonging to different behavior categories. The set of these probability values ​​can be called the behavior category probability array of the first point cloud feature sequence. The behavior category can include falling behavior, and at least one of squatting behavior, sitting behavior, bending behavior, climbing behavior or straight walking behavior. Correspondingly, the behavior category probability array can include the probability values ​​of each behavior category.

[0119] Taking the behavior categories including falling behavior, squatting behavior, sitting behavior, bending over behavior, climbing behavior, and walking straight behavior as an example, the corresponding behavior category probability array can be {p1, p2, ... p6}, where p1-p6 respectively represent the probabilities of the six behaviors, namely falling behavior, squatting behavior, sitting behavior, bending over behavior, climbing behavior, and walking straight behavior.

[0120] It should be noted that the conventional neural network model adopts a fully connected mode, that is, the point cloud features are expanded into a one-dimensional array as the input of the model, that is, the point cloud features of a single frame can include 7 inputs (the above-mentioned totalNum, xsize, ysize, zsize, center, ycenter, zcenter). The model only uses each point cloud feature as an independent input, and cannot well characterize the relative relationship between radar frames. However, when a person falls or does other movements, the size and coordinates of the front and rear radar frames have obvious time variation patterns. For example, the state judgment of a person's behavior at time t depends not only on the input features at time t, but also on the input features at time t-1. Therefore, a recurrent neural network model can be selected to describe the data changes between radar frames.

[0121] It is worth mentioning that the number of radar frames corresponding to the point cloud features to be recorded in the first point cloud feature sequence, that is, the first preset number mentioned above, can be determined in the following manner. The corresponding formula is as follows:

[0122] featureList={featureArr(1),featureArr(2)...featureArr(lastFrame)};

[0123]

[0124] Among them, lastFrame represents the first preset number, lastingTime represents the duration of the fall obtained by statistics, such as the average value of the duration of the fall obtained by big data statistics or the duration of the fall given according to the data change trend, such as 1.5-2 seconds, and Period represents the frame period of the radar.

[0125] Step S130: Generate a fall detection result according to the behavior category probability array of the first point cloud feature sequence.

[0126] In some embodiments, the behavior category corresponding to the maximum probability value in the behavior category probability array of the first point cloud feature sequence may be obtained. If the behavior category corresponding to the maximum probability value is a fall behavior, a fall detection result indicating the presence of a fall is generated, otherwise a fall detection result indicating the absence of a fall is generated.

[0127] Taking the above-mentioned behavior category probability array presentation form of {p1, p2, ... p6} as an example, if the behavior category probability array is {0.02, 0.02, 0.03, 0.9, 0.01, 0.02}, the maximum probability value is 0.9, and the corresponding behavior category is the above-mentioned p4, that is, the corresponding behavior category is bending behavior. Since bending behavior is not falling behavior, a fall detection result of no fall is generated.

[0128] It is worth mentioning that the fall detection results can also be judged again in combination with the probability threshold to improve the detection accuracy. Taking the probability threshold of 0.8 as an example, if the maximum probability value in the behavior category probability array is less than 0.8, the behavior category probability array can be considered to be an invalid array with little reference value. The corresponding behavior category can be directly determined as an invalid behavior. Since invalid behavior is not a fall behavior, a fall detection result without a fall can also be generated.

[0129] In some embodiments, considering that a single point cloud feature sequence has a large randomness, multiple point cloud feature sequences can be combined to generate fall detection results to improve the accuracy of fall detection. For example, a fall detection result can be generated based on the behavior category probability array of the first point cloud feature sequence and the behavior category probability array of at least one second point cloud feature sequence acquired in advance. Among them, at least one second point cloud feature sequence is an adjacent point cloud feature sequence acquired before the first point cloud feature sequence, and the adjacent point cloud feature sequence sequentially records the point cloud features of each of the first preset number of continuous radar frames, and the adjacent point cloud feature sequence and the first point cloud feature sequence acquired thereafter have the following characteristics: all radar frames corresponding to the second point cloud feature to the last point cloud feature recorded in sequence in the adjacent point cloud feature sequence are the same as all radar frames corresponding to the first point cloud feature to the second to last point cloud feature recorded in sequence in the first point cloud feature sequence acquired thereafter.

[0130] Specifically, first, the behavior category corresponding to the maximum probability value in the behavior category probability array of the target point cloud feature sequence can be determined as the behavior category of the target point cloud feature sequence, wherein the target point cloud feature sequence is any one of the first point cloud feature sequence and at least one second point cloud feature sequence. Afterwards, it is determined whether the proportion of the target point cloud feature sequences with the behavior category of falling behavior to all target point cloud feature sequences exceeds a preset threshold. If the proportion of the target point cloud feature sequences with the behavior category of falling behavior to all target point cloud feature sequences exceeds the preset threshold, a fall detection result indicating the presence of a fall is generated, otherwise a fall detection result indicating the absence of a fall is generated.

[0131] For example, classList can be used to represent all target point cloud feature sequences, namely:

[0132] classList={class(1),class(2)......class(saveFrame)}

[0133] Among them, saveFrame represents the number of all target point cloud feature sequences, and class(i) represents the behavior category corresponding to the maximum probability value in the i-th target point cloud feature sequence. In this way, the proportion of target point cloud feature sequences corresponding to fall behaviors in classList can be counted. When the proportion is greater than the preset threshold, it is considered that fall behaviors exist. In this way, the fall detection result with fall can be output.

[0134] In order to further improve the detection accuracy, a false alarm suppression mechanism based on actual logic can be combined to determine whether to output a fall alarm message.

[0135] Specifically, since temporal behaviors such as falling are most sensitive to height, false alarm suppression can be based on height changes. First, obtain the height value of the radar frame corresponding to the last point cloud feature recorded in sequence in the first point cloud feature sequence, and the height values ​​of the second preset number of radar frames obtained continuously before the radar frame corresponding to the last point cloud feature. Afterwards, determine the first radar frame where the maximum height value is located and the second radar frame where the minimum height value is located among all the height values. If the maximum height value is greater than the first preset height, the minimum height value is less than the second preset height, and the time difference between the first radar frame and the second radar frame is greater than the preset time difference, it is considered to meet the fall behavior, so the fall alarm information can be output. If one of the maximum height value, the minimum height value or the above time difference does not meet the above situation, it is considered to not meet the fall behavior, so the fall alarm information can be not output.

[0136] In some embodiments, the fall warning information may be at least one of a voice message, a short message, or a buzzer message.

[0137] The embodiment of the present invention provides a fall detection method, device and electronic device, firstly obtaining the point cloud features of the target to be detected in the point cloud set of each of the first preset number of radar frames, and obtaining a first point cloud feature sequence which sequentially records the point cloud features of each of the first preset number of radar frames. Then, the first point cloud feature sequence is input into a preset recurrent neural network model to obtain a behavior category probability array of the first point cloud feature sequence. Finally, a fall detection result is generated according to the behavior category probability array of the first point cloud feature sequence.

[0138] Since the point cloud features include the point count features, size features, and centroid features of the target to be detected, and cover information such as the height change and size of the human body to be detected, the point cloud features can more comprehensively describe the instantaneous state and process of a human body falling. In addition, the preset recurrent neural network model adopted is trained based on point cloud feature sequence samples of multiple behavior categories that are pre-labeled. It can well distinguish interfering movements such as squatting, sitting, and bending over from falls, so the accuracy of fall detection is high. In addition, the fall detection method provided in the embodiment of the present invention can be integrated on a millimeter wave radar, without the need for the user to carry it with them, and has the advantages of high ease of use.

[0139] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0140] The following is an embodiment of the device of the present invention. For details not described in detail therein, reference may be made to the corresponding method embodiment described above.

[0141] Figure 2 The structure diagram of the fall detection device provided by the embodiment of the present invention is shown. For the convenience of description, only the part related to the embodiment of the present invention is shown, which is described in detail as follows:

[0142] like Figure 2 As shown, the fall detection device 200 includes:

[0143] The acquisition module 210 is used to acquire point cloud features of the target to be detected in the point cloud sets of each of the first preset number of radar frames, and obtain a first point cloud feature sequence that sequentially records the point cloud features of each of the first preset number of radar frames; wherein the point cloud features include point count features, size features, and centroid features of the target to be detected;

[0144] An input module 220 is used to input the first point cloud feature sequence into a preset recurrent neural network model to obtain a behavior category probability array of the first point cloud feature sequence; wherein the preset recurrent neural network model is trained based on point cloud feature sequence samples of multiple behavior categories that are pre-labeled, and the behavior categories include falling behavior, and at least one of squatting behavior, sitting behavior, bending behavior, climbing behavior, or walking straight behavior, and the behavior category probability array includes probability values ​​of each behavior category;

[0145] The detection module 230 is used to generate a fall detection result according to the behavior category probability array of the first point cloud feature sequence.

[0146] In a possible implementation, the acquisition module is further used to:

[0147] Acquiring a first preset number of continuous radar frames;

[0148] The point cloud set of a single acquired radar frame is determined as the point cloud set of the corresponding radar frame; or, the point cloud set of the acquired target radar frame and the point cloud set of the last radar frame acquired before the target radar frame are mapped to the same coordinate system, and the mapped point cloud set is determined as the point cloud set of the target radar frame; wherein the target radar frame is any one of the first preset number of radar frames.

[0149] In a possible implementation, the acquisition module is further used to:

[0150] Determine all points corresponding to the target to be detected from the point cloud set;

[0151] The number of all points is determined as the point number feature of the target to be detected;

[0152] Obtain the maximum X-axis coordinate, the maximum Y-axis coordinate, the maximum Z-axis coordinate, the minimum X-axis coordinate, the minimum Y-axis coordinate, and the minimum Z-axis coordinate of all the point coordinates, and determine the difference between the maximum X-axis coordinate and the minimum X-axis coordinate, the difference between the maximum Y-axis coordinate and the minimum Y-axis coordinate, and the difference between the maximum Z-axis coordinate and the minimum Z-axis coordinate as the size feature of the target to be detected;

[0153] Accumulate the first product value of the X coordinate of each point and the signal-to-noise ratio, the second product value of the Y coordinate of each point and the signal-to-noise ratio, and the third product value of the Z coordinate of each point and the signal-to-noise ratio respectively, and calculate the first quotient value of the sum of the accumulated first product values ​​and the number of all points, the second quotient value of the sum of the second product values ​​and the number of all points, and the third quotient value of the sum of the third product values ​​and the number of all points respectively, and determine the first quotient value, the second quotient value, and the third quotient value as the centroid feature of the target to be detected.

[0154] In a possible implementation, the fall detection system further includes a normalization module, which is used to:

[0155] Normalizing the first point cloud feature sequence to obtain a normalized first point cloud feature sequence;

[0156] Input the first point cloud feature sequence into a preset recurrent neural network model, including:

[0157] The normalized first point cloud feature sequence is input into the preset recurrent neural network model.

[0158] In a possible implementation, the detection module is further used to:

[0159] Obtain the behavior category corresponding to the maximum probability value in the behavior category probability array of the first point cloud feature sequence;

[0160] If the behavior category corresponding to the maximum probability value is a fall behavior, a fall detection result indicating that a fall exists is generated; otherwise, a fall detection result indicating that a fall does not exist is generated.

[0161] In a possible implementation, the detection module is further used to:

[0162] Generate a fall detection result according to the behavior category probability array of the first point cloud feature sequence and the behavior category probability array of at least one second point cloud feature sequence acquired in advance;

[0163] Among them, at least one second point cloud feature sequence is an adjacent point cloud feature sequence acquired before the first point cloud feature sequence, the adjacent point cloud feature sequence sequentially records the point cloud features of each of a first preset number of consecutive radar frames, and the adjacent point cloud feature sequence and the point cloud feature sequence first acquired thereafter have the following characteristics: all radar frames corresponding to the second point cloud feature to the last point cloud feature recorded sequentially in the adjacent point cloud feature sequence are the same as all radar frames corresponding to the first point cloud feature to the second to last point cloud feature recorded sequentially in the point cloud feature sequence first acquired thereafter.

[0164] In a possible implementation, the detection module is further used to:

[0165] Determine the behavior category corresponding to the maximum probability value in the behavior category probability array of the target point cloud feature sequence as the behavior category of the target point cloud feature sequence; wherein the target point cloud feature sequence is any one of the first point cloud feature sequence and at least one second point cloud feature sequence;

[0166] If the proportion of the target point cloud feature sequences whose behavior category is fall behavior to all target point cloud feature sequences exceeds a preset threshold, a fall detection result indicating the presence of a fall is generated; otherwise, a fall detection result indicating the absence of a fall is generated.

[0167] In a possible implementation, the fall detection system further includes an output module for:

[0168] Acquire a height value of a radar frame corresponding to the last point cloud feature recorded in sequence in the first point cloud feature sequence, and a height value of each of a second preset number of radar frames acquired successively before the radar frame corresponding to the last point cloud feature;

[0169] Determine, among all the altitude values, a first radar frame where the maximum altitude value is located and a second radar frame where the minimum altitude value is located;

[0170] If the maximum height value is greater than the first preset height, the minimum height value is less than the second preset height, and the time difference between the first radar frame and the second radar frame is greater than the preset time difference, a fall alarm message is output.

[0171] The embodiment of the present invention provides a fall detection method, device and electronic device, firstly obtaining the point cloud features of the target to be detected in the point cloud set of each of the first preset number of radar frames, and obtaining a first point cloud feature sequence which sequentially records the point cloud features of each of the first preset number of radar frames. Then, the first point cloud feature sequence is input into a preset recurrent neural network model to obtain a behavior category probability array of the first point cloud feature sequence. Finally, a fall detection result is generated according to the behavior category probability array of the first point cloud feature sequence.

[0172] Since the point cloud features include the point count features, size features, and centroid features of the target to be detected, and cover information such as the height change and size of the human body to be detected, the point cloud features can more comprehensively describe the instantaneous state and process of a human body falling. In addition, the preset recurrent neural network model adopted is trained based on point cloud feature sequence samples of multiple behavior categories that are pre-labeled. It can well distinguish interfering movements such as squatting, sitting, and bending over from falls, so the accuracy of fall detection is high. In addition, the fall detection method provided in the embodiment of the present invention can be integrated on a millimeter wave radar, without the need for the user to carry it with them, and has the advantages of high ease of use.

[0173] Figure 3 Schematic diagram of an electronic device 3 provided in an embodiment of the present invention. Figure 3 As shown, the electronic device 3 of this embodiment includes: a processor 30, a memory 31, and a computer program 32 stored in the memory 31 and executable on the processor 30. When the processor 30 executes the computer program 32, the steps in the above-mentioned various fall detection method embodiments are implemented, for example Figure 1 Alternatively, when the processor 30 executes the computer program 32, the functions of each module in the above-mentioned device embodiments are realized, for example, Figure 2 The functions of the modules 210 to 230 are shown.

[0174] Exemplarily, the computer program 32 may be divided into one or more modules, which are stored in the memory 31 and executed by the processor 30 to implement the present invention. The one or more modules may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program 32 in the electronic device 3. For example, the computer program 32 may be divided into Figure 2 Modules 210 to 230 are shown.

[0175] The electronic device 3 may include, but is not limited to, a processor 30 and a memory 31. Those skilled in the art will appreciate that Figure 3 It is only an example of the electronic device 3 and does not constitute a limitation of the electronic device 3. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device may also include input and output devices, network access devices, buses, etc.

[0176] The processor 30 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0177] The memory 31 may be an internal storage unit of the electronic device 3, such as a hard disk or memory of the electronic device 3. The memory 31 may also be an external storage device of the electronic device 3, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 3. Further, the memory 31 may also include both an internal storage unit and an external storage device of the electronic device 3. The memory 31 is used to store the computer program and other programs and data required by the electronic device. The memory 31 may also be used to temporarily store data that has been output or is to be output.

[0178] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0179] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0180] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0181] In the embodiments provided by the present invention, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0182] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0183] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0184] If the integrated module is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned fall detection method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0185] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A fall detection method, characterized in that: include: Acquire point cloud features of the target to be detected in point cloud sets of each of a first preset number of continuous radar frames, and obtain a first point cloud feature sequence that sequentially records the point cloud features of each of the first preset number of radar frames; Inputting the first point cloud feature sequence into a preset recurrent neural network model to obtain a behavior category probability array of the first point cloud feature sequence; wherein the preset recurrent neural network model is trained based on point cloud feature sequence samples of multiple behavior categories that are pre-labeled, the behavior categories include falling behavior, and at least one of squatting behavior, sitting behavior, bending behavior, climbing behavior, or walking straight behavior, and the behavior category probability array includes probability values ​​of each behavior category; generating a fall detection result according to the behavior category probability array of the first point cloud feature sequence; The point cloud features include point count features, size features, and centroid features of the target to be detected; and the step of obtaining the point cloud features of the target to be detected in the point cloud sets of each of the first preset number of consecutive radar frames includes: Determine all points corresponding to the target to be detected from the point cloud set; Determine the number of all points as the point number feature of the target to be detected; Obtaining the maximum X-axis coordinate, the maximum Y-axis coordinate, the maximum Z-axis coordinate, the minimum X-axis coordinate, the minimum Y-axis coordinate, and the minimum Z-axis coordinate of the point coordinates of all the points, and determining the difference between the maximum X-axis coordinate and the minimum X-axis coordinate, the difference between the maximum Y-axis coordinate and the minimum Y-axis coordinate, and the difference between the maximum Z-axis coordinate and the minimum Z-axis coordinate as the size feature of the target to be detected; Accumulate the first product value of the X coordinate of each point and the signal-to-noise ratio, the second product value of the Y coordinate of each point and the signal-to-noise ratio, and the third product value of the Z coordinate of each point and the signal-to-noise ratio respectively, and calculate the first quotient value of the sum of the accumulated first product values ​​and the number of all points, the second quotient value of the sum of the second product values ​​and the number of all points, and the third quotient value of the sum of the third product values ​​and the number of all points respectively, and determine the first quotient value, the second quotient value, and the third quotient value as the centroid feature of the target to be detected.

2. The method according to claim 1, characterized in that Before acquiring point cloud features of the target to be detected in the point cloud sets of each of the first preset number of consecutive radar frames, the method further includes: Acquiring a first preset number of continuous radar frames; The point cloud set of a single acquired radar frame is determined as the point cloud set of the corresponding radar frame; or, the point cloud set of the acquired target radar frame and the point cloud set of the last radar frame acquired before the target radar frame are mapped to the same coordinate system, and the mapped point cloud set is determined as the point cloud set of the target radar frame; wherein the target radar frame is any one of the first preset number of radar frames.

3. The method according to claim 1, characterized in that Before inputting the first point cloud feature sequence into a preset recurrent neural network model, the method further includes: Normalizing the first point cloud feature sequence to obtain a normalized first point cloud feature sequence; The step of inputting the first point cloud feature sequence into a preset recurrent neural network model comprises: The normalized first point cloud feature sequence is input into a preset recurrent neural network model.

4. The method according to claim 1, characterized in that: The generating of the fall detection result according to the behavior category probability array of the first point cloud feature sequence comprises: Obtaining the behavior category corresponding to the maximum probability value in the behavior category probability array of the first point cloud feature sequence; If the behavior category corresponding to the maximum probability value is a fall behavior, a fall detection result indicating that a fall exists is generated; otherwise, a fall detection result indicating that a fall does not exist is generated.

5. The method according to claim 1, characterized in that The generating of the fall detection result according to the behavior category probability array of the first point cloud feature sequence comprises: generating a fall detection result according to the behavior category probability array of the first point cloud feature sequence and the behavior category probability array of at least one second point cloud feature sequence acquired in advance; Among them, the at least one second point cloud feature sequence is an adjacent point cloud feature sequence acquired before the first point cloud feature sequence, and the adjacent point cloud feature sequence sequentially records the point cloud features of each of the first preset number of consecutive radar frames, and the adjacent point cloud feature sequence and the point cloud feature sequence first acquired thereafter have the following characteristics: all radar frames corresponding to the second point cloud feature to the last point cloud feature recorded sequentially in the adjacent point cloud feature sequence are the same as all radar frames corresponding to the first point cloud feature to the second to last point cloud feature recorded sequentially in the point cloud feature sequence first acquired thereafter.

6. The method according to claim 5, characterized in that The generating of the fall detection result according to the behavior category probability array of the first point cloud feature sequence and the behavior category probability array of at least one second point cloud feature sequence acquired in advance comprises: Determine the behavior category corresponding to the maximum probability value in the behavior category probability array of the target point cloud feature sequence as the behavior category of the target point cloud feature sequence; wherein the target point cloud feature sequence is any one of the first point cloud feature sequence and the at least one second point cloud feature sequence; If the proportion of the target point cloud feature sequences whose behavior category is fall behavior to all the target point cloud feature sequences exceeds a preset threshold, a fall detection result indicating the presence of a fall is generated; otherwise, a fall detection result indicating the absence of a fall is generated.

7. The method according to claim 1, characterized in that After generating the fall detection result, the method further includes: Acquire a height value of a radar frame corresponding to the last point cloud feature recorded in sequence in the first point cloud feature sequence, and a height value of each of a second preset number of radar frames acquired successively before the radar frame corresponding to the last point cloud feature; Determine, among all the altitude values, a first radar frame where a maximum altitude value is located and a second radar frame where a minimum altitude value is located; If the maximum height value is greater than a first preset height, the minimum height value is less than a second preset height, and the time difference between the first radar frame and the second radar frame is greater than a preset time difference, a fall alarm message is output.

8. A fall detection device, characterized in that: include: An acquisition module is used to acquire point cloud features of the target to be detected in the point cloud sets of each of the first preset number of radar frames, and obtain a first point cloud feature sequence that sequentially records the point cloud features of each of the first preset number of radar frames; wherein the point cloud features include point number features, size features, and centroid features of the target to be detected; An input module, used for inputting the first point cloud feature sequence into a preset recurrent neural network model to obtain a behavior category probability array of the first point cloud feature sequence; wherein the preset recurrent neural network model is trained based on point cloud feature sequence samples of multiple behavior categories that are pre-labeled, the behavior categories include falling behavior, and at least one of squatting behavior, sitting behavior, bending behavior, climbing behavior or walking straight behavior, and the behavior category probability array includes probability values ​​of each behavior category; A detection module, configured to generate a fall detection result according to the behavior category probability array of the first point cloud feature sequence; The point cloud features include point count features, size features and centroid features of the target to be detected; the acquisition module is also used to determine all points corresponding to the target to be detected from the point cloud set; Determine the number of all points as the point number feature of the target to be detected; Obtaining the maximum X-axis coordinate, the maximum Y-axis coordinate, the maximum Z-axis coordinate, the minimum X-axis coordinate, the minimum Y-axis coordinate, and the minimum Z-axis coordinate of the point coordinates of all the points, and determining the difference between the maximum X-axis coordinate and the minimum X-axis coordinate, the difference between the maximum Y-axis coordinate and the minimum Y-axis coordinate, and the difference between the maximum Z-axis coordinate and the minimum Z-axis coordinate as the size feature of the target to be detected; Accumulate the first product value of the X coordinate of each point and the signal-to-noise ratio, the second product value of the Y coordinate of each point and the signal-to-noise ratio, and the third product value of the Z coordinate of each point and the signal-to-noise ratio respectively, and calculate the first quotient value of the sum of the accumulated first product values ​​and the number of all points, the second quotient value of the sum of the second product values ​​and the number of all points, and the third quotient value of the sum of the third product values ​​and the number of all points respectively, and determine the first quotient value, the second quotient value, and the third quotient value as the centroid feature of the target to be detected.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

Images