Fall detection intelligent alarm system based on millimeter wave radar

Abstract

The present invention provides a fall detection intelligent alarm system based on millimeter wave radar, which divides the historical data set for a specific human body into a fall subset and a non-fall subset according to the previous "fall" and "non-fall" states, and analyzes the fall subset and the non-fall subset separately. Only after the respective judgments of the two subsets can it be jointly determined whether the specific human body has fallen, and thus the fall signal is sent to the remote server for alarm. The present invention is aimed at the historical data of the specific human body, so it fully achieves personalized fall alarm according to the different habits of the individual, and on this basis, the fall conclusion is drawn after two judgments, thereby preventing the risk of missed reports and false reports to the greatest extent.

Description

Technical Field

[0001] The invention relates to a fall detection intelligent alarm system based on millimeter wave radar. Background Art

[0002] In recent years, as my country gradually enters an aging society, the health problems of the elderly have received more and more attention from the society. For many elderly people, a minor fall may cause serious and irreversible physiological damage if the police are not notified in time and rescue is not timely.

[0003] Therefore, various types of fall alarm devices have emerged in the market, which can quickly determine whether the elderly have fallen by sensing their physical condition, and quickly determine whether a fall alarm is needed based on this judgment. However, such devices are often general-purpose devices, which take into account the commonalities of the elderly group, but often ignore the differences among the elderly individuals, resulting in a large number of false alarms and missed alarms.

[0004] For example, some instruments use knee height as a sensing indicator, and believe that if the elderly's knees are momentarily below a certain height, the instrument will conclude that the elderly have fallen. This is of course reasonable, because falls are often accompanied by a momentary decrease in knee height, which is a highly probable and relatively logical idea. However, if this is used for fall alarms, it is likely that a large number of false alarms will occur for many elderly people. Because many elderly people often perform knee bending exercises to exercise, the knees will frequently lower during the knee bending exercise, resulting in a large number of false alarms of falls. Summary of the invention

[0005] The present invention provides a fall detection intelligent alarm system based on millimeter wave radar, which effectively solves the above-mentioned technical problems existing in the prior art.

[0006] Specifically, the present invention provides a fall detection intelligent alarm system based on millimeter wave radar, the system includes n body sensors, a central processing unit, and a remote server, wherein the n body sensors are closely attached to multiple parts of a specific human body to respectively sense n kinds of real-time body indicators A1, A2, ...An or a1, a2, ...an of the specific human body, wherein A1, A2, ...An and a1, a2, ...an are one-to-one equivalent, and the n kinds of real-time body indicators are sent to the central processing unit at each specific moment, thereby the central processing unit stores a historical data set, the historical data set includes a total of k moments of real-time body indicators and the fall state corresponding to each moment, that is, "fall" and "not fall", in the central processing unit, the historical data set is divided into a fall data subset and a non-fall data subset according to the two different states of "fall / not fall", and the central processing unit first performs a fall judgment, wherein the fall indicator average value M' of each real-time body indicator Aj in the fall data subset is calculated j , then, based on the average value M' j Calculate the variance T' of each instant shape index Aj in this subset j , that is, calculate the variance T'1, T'2, ..., T' for n instant body indicators n , where the integer j ranges from 1 to n, with variances T'1, T'2, ..., T' n After sorting by numerical value, select the N variances with the smallest values, where 1≤N≤n, and the N variances correspond to the respective fall index average values ​​M1, ..., M N , the specific human body at the current moment each body sensor detects N real-time body indicators A'1, ..., A' in the fall judgment N The actual sensing value is P1, ....P N , based on the average values ​​of N fall indicators and N real sensing values, the fall distance L is calculated, and the fall threshold MaxL is set. If the fall distance L≤MaxL, the specific human body is determined to be in a pending fall state. If the fall distance L>MaxL, it is directly determined that the specific human body has not fallen. Then, the central processing unit performs the non-fall judgment and calculates the non-fall indicator average value m' of each instant body indicator aj in the non-fall data subset j , then, based on the average value m' j Find the variance t' of each instant shape index aj in this subset j , that is, calculate the variance t'1, t'2, ..., t' for n instant body indicators n , where the integer j ranges from 1 to n, with variances t'1, t'2, ..., t' n After sorting by numerical value, select the N variances with the smallest values. These N variances correspond to the respective uninverted index average values ​​m1, ..., mN , the specific human body at the current moment each body sensor detects N real-time body indicators a'1, ..., a' in the fall judgment N The actual sensing value is p1, ....p N , the non-falling distance l is calculated based on the average value of the non-falling index and the N real sensing values ​​in the non-falling judgment, and the non-falling threshold Maxl is set. If the non-falling distance l≤Maxl, the specific human body originally in the pending falling state is judged to be in the non-falling state, and if the non-falling distance l>Maxl, the specific human body is finally judged to be in the falling state. Once the central processing unit determines that the specific human body is in the falling state, the "fall warning signal" is sent to the remote server, and once it is determined that the specific human body is in the non-falling state, the "non-falling signal" is sent to the remote server.

[0007] Preferably, the fall data subset includes data at k' moments, and the data at any i'th moment among the k' moments is [A 1i’ , A 2i’ , …, A ni’ , fall], where i' is taken from 1 to k'; the unfallen data subset includes data at k'' moments, and the data at any i'' moment among the k'' moments is [a 1i” 、a 2i ”、…、a ni” , not inverted], where i” takes values ​​from 1 to k”.

[0008] Preferably, in the fall judgment, the average value M' of any instantaneous body index Aj in the fall data subset is j Calculated as:

[0009]

[0010] In the above formula, the subscript of A contains two digits, the first digit represents the type number of the real-time body index, the second digit represents the sensing time number of the real-time body index, and then, based on the average value M' j Calculate the variance T' of the instant shape index Aj in this subset j :

[0011]

[0012] Thus, the variances T'1, T'2, ..., T' of n instant body indicators A1, A2, ..., An in the fall data subset are calculated. n , in the variance T'1, T'2, ..., T' n After sorting by numerical value, select the N variances with the smallest values, where 1≤N≤n, and the N fall instantaneous body indicators corresponding to them in the n instantaneous body indicators are counted as A'1, ..., A'N , one by one corresponding to the respective fall index average values ​​M1, ..., M N .

[0013] Preferably, the fall distance L is calculated based on the fall index average value and N sensing values:

[0014]

[0015] The integer x in the above equation takes values ​​from 1 to N.

[0016] Preferably, in the non-falling judgment, the non-falling index average m' of any instant body index aj j The calculation formula is as follows:

[0017] Among them, the integer j is traversed from to n.

[0018] Preferably, based on the average value m' j Find the variance t' of the instantaneous shape index aj in this subset j :

[0019]

[0020] Thus, the variances t'1, t'2, ..., t' of the n instant shape indicators a1, a2, ..., an in the uninverted data subset are calculated. n , in the variance t'1, t'2, ..., t' n After sorting by numerical value, select the N variances with the smallest values, and the N fall instantaneous body indicators corresponding to them in the n instantaneous body indicators are counted as a'1, ..., a' N , one by one corresponding to the average values ​​of the unreversed indicators m1, ..., m N .

[0021] Preferably, based on the fall index average value and N real sensing values ​​p1, ....p N Calculate the uninverted distance l:

[0022]

[0023] The integer y in the above equation takes values ​​from 1 to N.

[0024] In summary, the fall detection intelligent alarm system provided by the present invention divides the historical data set for a specific human body into a fall subset and a non-fall subset according to the previous "fall" and "non-fall" states, and analyzes the fall subset and the non-fall subset separately. Only after the two subsets are judged separately can it be jointly determined whether the specific human body has fallen, and thus the fall signal is sent to the remote server for alarm. The present invention is aimed at the historical data of a specific human body, so it fully realizes personalized fall alarm according to different personal habits. On this basis, the fall conclusion will be drawn after two judgments, thus minimizing the risk of missed reports and false reports. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be discussed below. Obviously, the technical solutions described in conjunction with the drawings are only some embodiments of the present invention. For ordinary technicians in this field, other embodiments and their drawings can be obtained based on the embodiments shown in these drawings without paying creative work.

[0026] Figure 1 The specific operation flow chart of the millimeter wave radar-based fall detection intelligent alarm system according to the present invention is shown. DETAILED DESCRIPTION

[0027] The technical solutions of various embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments described in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] In summary, the present invention provides a fall detection intelligent alarm system, which makes full use of the historical data of a specific human body, analyzes the alarm data and the non-alarm data separately, and comprehensively judges whether the specific human body has fallen from both positive indicators and negative indicators based on the analysis results and the current movement state of the human body, and further decides whether to issue an alarm signal based on the judgment result.

[0029] In the following, we will base on Figure 1 The figure describes in detail the specific operation process of the fall detection intelligent alarm system provided by the present invention. Figure 1 The specific operation flow chart of the millimeter wave radar-based fall detection intelligent alarm system according to the present invention is shown.

[0030] The fall detection intelligent alarm system provided by the present invention comprises n body sensors (n≥2, in actual situations, n can usually take values ​​of 2, 3, 4, 5, indicating that the number of body sensors is usually between 2-5), a central processing unit, and a remote server. The n body sensors are closely attached to multiple parts of a specific human body to respectively sense n kinds of real-time body indicators of the specific human body. The central processing unit obtains the n kinds of real-time body indicators sensed by sending millimeter-wave radars. The central processing unit is responsible for judging whether the specific human body falls according to the n kinds of real-time body indicators. When judging that the specific human body falls, the alarm signal is remotely sent to the remote server.

[0031] The central processing unit can be loaded into a smart phone or tablet computer in the form of APP software, for example.

[0032] In addition, for example, the so-called n kinds of instant body indicators sensed by n body sensors can specifically refer to knee height, knee bending angular velocity, head height, abdomen height, and upper arm rotation speed. These five kinds of instant body indicators can cover the indicator characteristics that appear when a person falls in this example. For example, when a person falls, the knee height may drop instantly, or the knee may bend instantly, or it may be accompanied by a sudden drop in head height, or a sudden drop in abdomen height, or it may be accompanied by upper arm swinging. Then correspondingly, these five kinds of instant body indicators correspond to five kinds of body sensors, namely, a knee height sensor for sensing knee height, a knee bending angular velocity sensor for sensing knee bending angular velocity, a head height sensor for sensing head height, an abdomen height sensor for sensing abdomen height, and an upper arm rotation speed sensor for sensing upper arm rotation speed.

[0033] Each of the n body sensors historically senses a real-time body index once every specific time period (for example, every 1 minute) and stores it in the central processing unit. The central processing unit also stores the fall status corresponding to each real-time body index sensed (i.e., the two states of "falling" or "not falling").

[0034] In the following, A1, A2, ...A n A refers to the above n real-time body indicators. Further, if these n real-time body indicators are recorded at any i-th moment, then the n real-time body indicators at the i-th moment are recorded as A. 1i , A 2i , …A ni For the specific j-th body shape index Aj, it is recorded as follows from the first moment to the j-th moment: A j1 , A j2 , …A jkIn other words, in the following description, the subscript of A contains two numbers, the first number indicates the category of the real-time body indicator, and the second number indicates the time when the real-time body indicator is located.

[0035] It is assumed that the central processing unit stores a total of k instantaneous body indicators and the fall states corresponding to the instantaneous moments. Among them, the integer k is required to be at least k>3. In actual situations, k can often take hundreds or thousands of values, for example, k>1000, so that the inspection sample is large enough and as time goes by, the number of instantaneous body indicators stored in the historical database increases, and k will gradually increase.

[0036] In other words, at the i-th moment, the n instantaneous shape indicators are A 1i , A 2i , …, A ni , the corresponding falling state is "falling". Therefore, the i-th moment data stored in the central processing unit is [A 1i , A 2i , …, A ni Of course, if the corresponding falling state is “not falling”, the i-th moment data stored in the central processing unit is [A 1i , A 2i , …, A ni , not fallen].

[0037] Therefore, the above historical data set stored in the central processing unit can be divided into two subsets according to the two different states of "falling / not falling", namely the falling data subset and the not falling data subset.

[0038] The fall data subset includes data at k' moments, and the data at any i'th moment among the k' moments is [A 1i’ , A 2i’ , ..., A ni’ , fall], where i' takes values ​​from 1 to k'.

[0039] The uninverted data subset includes data at k" moments, and the data at any i"th moment among the k" moments is [a 1i” 、a 2i” , ..., a ni” , not inverted], where i” ranges from 1 to k. It should be noted that in order to distinguish the two subsets, when expressing n kinds of instant body indicators in each subset, A1, A2, ...A n and a1, a2, ...a n , but this is only to distinguish the respective operations in the following text. In fact, for any integer i in the interval [1,n], Ai and ai represent the same real-time shape indicator.

[0040] Obviously, here k=k'+k".

[0041] For example, the data set of the central processing unit can be listed as follows:

[0042] time A1 A2 A3 A4 Falling state 1 0.56 20 5.6 13 Not fallen 2 0.43 25 4.2 15 Not fallen 3 0.66 40 4.6 14 fall 4 0.61 34 4.2 12 Not fallen 5 0.53 24 3.2 16 fall 6 0.45 23 3.7 13 Not fallen 7 0.31 36 4.7 14 Not fallen 8 0.44 32 5.8 15 fall 9 0.21 45 3.2 17 Not fallen 10 0.45 35 7.2 14 fall

[0043] It should be noted that the numbers filled in the above table ignore the dimension (i.e., unit). For example, A2 at the first moment can represent the knee height of 20 cm, so the unit "cm" is ignored in the table. A1 can represent the angular velocity of the knee joint, so the unit can be radians / second, which is also ignored here.

[0044] In the above table, data at 10 moments in time are collected, so k=10.

[0045] The data set in the above table is divided into two categories: the falling data subset and the non-falling data subset. The falling data subset includes data at 4 moments, so k'=4, and the non-falling data subset includes data at 6 moments, so k"=6.

[0046] Next, firstly, the fall determination is performed. In the fall data subset, the central processing unit performs corresponding calculations. Firstly, the performance of the instant body index A1 in the fall data subset is examined.

[0047] First, calculate the fall index average value M'1 of the real-time body index A1 in the fall data subset:

[0048]

[0049] Then, based on the mean value M'1, the variance T'1 of the instant body index A1 in the subset is calculated:

[0050]

[0051] The integer z in the above equation takes values ​​from 1 to k'.

[0052] By analogy, the average value M' of any instantaneous body index Aj in the fall data subset can be calculated: j :

[0053]

[0054] Then, based on the average value M' j Calculate the variance T' of the instant shape index Aj in this subset j :

[0055]

[0056] Thus, the variances T'1, T'2, ..., T' of n instant body indicators A1, A2, ..., An in the fall data subset are calculated. n .

[0057] If the above-mentioned variance of a real-time body shape index is small (ie, relatively close to its average value), it means that the change of the real-time body shape index is small and is highly correlated with the fall state near the average value.

[0058] Therefore, in the variance T'1, T'2, ..., T' n After sorting by numerical value, select the N variances with the smallest values, where 1≤N≤n, and the N fall instantaneous body indicators corresponding to them in the n instantaneous body indicators are counted as A'1, ..., A' N , one by one corresponding to the respective fall index average values ​​M1, ..., M N .

[0059] In this case, the N instantaneous body indicators A'1, ..., A' detected by the body sensors of the specific human body at the current moment are N The actual sensing value is P1, ....P N .

[0060] Next, the fall distance L is calculated based on the average fall index and N fall real-time shape indicators:

[0061]

[0062] The integer x in the above equation takes values ​​from 1 to N.

[0063] Set a fall threshold MaxL. If the above fall distance L≤MaxL, it is determined that the specific human body is in a pending fall state, and then enters the non-fall judgment to determine whether it has fallen. If the fall distance L>MaxL, it is directly determined that the specific human body has not fallen.

[0064] Next, the non-falling judgment is introduced. In the non-falling data subset, the central processing unit also performs corresponding calculations. First, the performance of the instant body index A1 in the non-falling data subset is examined.

[0065] First, calculate the uninverted index average m'1 of the instant shape index a1 in the uninverted data subset:

[0066]

[0067] Then, based on the mean value m'1, the variance t'1 of the instantaneous body index a1 in the subset is calculated:

[0068]

[0069] The integer z in the above equation takes values ​​from 1 to k'.

[0070] By analogy, the average value m' of any instant shape index aj in the uninverted data subset can be calculated. j :

[0071]

[0072] Then, based on the average value m' j Find the variance t' of the instantaneous shape index aj in this subset j :

[0073]

[0074] Thus, the variances t'1, t'2, ..., t' of the n instant shape indicators a1, a2, ..., an in the uninverted data subset are calculated. n .

[0075] If the above-mentioned variance of an instant body shape index is small (i.e., relatively close to its average value), it means that the instant body shape index changes little and is highly correlated with the non-falling state near the average value.

[0076] Therefore, in the variance t'1, t'2, ..., t' n After sorting by numerical value, select the N variances with the smallest values, and the N fall instantaneous body indicators corresponding to them in the n instantaneous body indicators are counted as a'1, ..., a' N , one by one corresponding to the average values ​​of the unreversed indicators m1, ..., m N .

[0077] As described above, the N instantaneous body indicators a'1, ..., a' detected by the body sensors of a specific human body at the current moment are N The actual sensing value is p1, ....p N .

[0078] Next, based on the average value of the fall indicator and N real sensing values ​​p1, ....p N Calculate the uninverted distance l:

[0079]

[0080] The integer y in the above equation takes values ​​from 1 to N.

[0081] A non-falling threshold Maxl is set. If the non-falling distance l≤Maxl, the specific human body originally in the pending falling state is finally determined to be in the non-falling state. If the non-falling distance l>Maxl, the specific human body is finally determined to be in the falling state.

[0082] It can be seen that to finally determine that a specific human body is in a falling state, two determinations are required, namely the falling determination and the not falling determination. If it is determined as "not falling" in the falling determination, it will be directly determined as "not falling" in the final determination. If it is determined as "pending falling state" in the falling determination, it needs to enter the not falling determination as a "candidate". Only when it is determined as "falling state" in the not falling determination, the system will finally confirm that the specific human body is in a falling state.

[0083] Once the central processing unit determines that the specific human body is in a falling state, it sends a "fall warning signal" to the remote server, and once it determines that the specific human body is in a non-falling state, it sends a "non-falling signal" to the remote server.

[0084] At this point, the fall detection intelligent alarm system provided by the present invention has been basically introduced. In the fall detection intelligent alarm system of the present invention, the historical data set for a specific human body is divided into a fall subset and a non-fall subset according to the previous "fall" and "non-fall" states, and the fall subset and the non-fall subset are analyzed separately. Only after the respective judgments of the two subsets can it be jointly determined whether the specific human body has fallen, and thus the fall signal is sent to the remote server for alarm. The present invention is aimed at the historical data of a specific human body, so it fully achieves personalized fall alarms according to different personal habits. On this basis, the fall conclusion is drawn after two judgments, thereby minimizing the risk of missed reports and false alarms.

[0085] The above description is only an exemplary embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A fall detection intelligent alarm system based on millimeter wave radar, characterized in that: The system includes n body sensors, a central processing unit, and a remote server. The n body sensors are closely attached to multiple parts of a specific human body to respectively sense n kinds of real-time body indicators A1, A2, ...An or a1, a2, ...an of the specific human body, wherein A1, A2, ...An and a1, a2, ...an are one-to-one equivalent. The central processing unit obtains the n kinds of real-time body indicators through a millimeter-wave radar at each specific moment, thereby storing a historical data set in the central processing unit. The historical data set includes a total of k moments of real-time body indicators and a falling state corresponding to each moment, that is, "falling" and "not falling". In the central processing unit, the historical data set is divided into a falling data subset and a not falling data subset according to the two different states of "falling / not falling". The central processing unit first performs fall determination, wherein the fall index average M' of each instantaneous body index Aj in the fall data subset is calculated j , then, based on the average value M' j Calculate the variance T' of each instant shape index Aj in this subset j , that is, calculate the variance T'1, T'2, ..., T' for n instant body indicators n , where the integer j ranges from 1 to n, In the variance T'1, T'2, ..., T' n After sorting by numerical value, select the N variances with the smallest values, where 1≤N≤n, and the N variances correspond to the respective fall index average values ​​M1, ..., M N , the specific human body at the current moment each body sensor detects N real-time body indicators A'1, ..., A' in the fall judgment N The actual sensing value is P1, ....P N , The falling distance L is calculated based on the average values ​​of N falling indicators and N real sensing values, and the falling threshold MaxL is set. If the falling distance L≤MaxL, it is determined that the specific human body is in a pending falling state. If the falling distance L>MaxL, it is directly determined that the specific human body has not fallen. Next, the central processing unit performs the non-falling judgment and calculates the non-falling index average m' of each instant shape index aj in the non-falling data subset. j , then, based on the average value m' j Find the variance t' of each instant shape index aj in this subset j , that is, calculate the variance t'1, t'2, ..., t' for n instant body indicators n , where the integer j ranges from 1 to n, In the variance t'1, t'2, ..., t' n After sorting by numerical value, select the N variances with the smallest values. These N variances correspond to the respective uninverted index average values ​​m1, ..., m N , At the current moment, each body sensor detects N real-time body indicators a'1, ..., a' in the fall judgment of a specific human body. N The actual sensing value is p1, ....p N , The non-falling distance l is calculated based on the average value of the non-falling index and the N real sensing values ​​in the non-falling judgment, and the non-falling threshold Maxl is set. If the non-falling distance l≤Maxl, the specific human body originally in the pending falling state is determined to be in the non-falling state, and if the non-falling distance l>Maxl, the specific human body is finally determined to be in the falling state. Once the central processing unit determines that a specific human body is in a falling state, it sends a "fall warning signal" to the remote server, and once it determines that the specific human body is not in a falling state, it sends a "non-fall signal" to the remote server. The fall data subset includes data at k' moments, and the data at any i'th moment among the k' moments is [A 1i’ , A 2i’ , …, A ni’ , fall], where i' is taken from 1 to k'; the unfallen data subset includes data at k'' moments, and the data at any i'' moment among the k'' moments is [a 1i” 、a 2i” , …, a ni” , not inverted], where i” takes values ​​from 1 to k, In the fall judgment, the average value M' of any instantaneous body index Aj in the fall data subset j Calculated as: In the above formula, the subscript of A contains two digits, the first digit represents the type number of the real-time body indicator, and the second digit represents the sensing time number of the real-time body indicator. Then, based on the average value M' j Calculate the variance T' of the instant shape index Aj in this subset j : Thus, the variances T'1, T'2, ..., T' of n instant body indicators A1, A2, ..., An in the fall data subset are calculated. n , in the variance T'1, T'2, ..., T' n After sorting by numerical value, select the N variances with the smallest values, where 1≤N≤n, and the N fall instantaneous body indicators corresponding to them in the n instantaneous body indicators are counted as A'1, ..., A' N , one by one corresponding to the respective fall index average values ​​M1, ..., M N , The fall distance L is calculated based on the fall index average value and N sensing values: The integer x in the above equation takes values ​​from 1 to N. In the non-falling judgment, the non-falling index average m' of any instant shape index aj j The calculation formula is as follows: Among them, the integer j is traversed from to n, Based on the average value m' j Find the variance t' of the instantaneous shape index aj in this subset j : Thus, the variances t'1, t'2, ..., t' of the n instant shape indicators a1, a2, ..., an in the uninverted data subset are calculated. n , in the variance t'1, t'2, ..., t' n After sorting by numerical value, select the N variances with the smallest values, and the N fall instantaneous body indicators corresponding to them in the n instantaneous body indicators are counted as a'1, ..., a' N , one by one corresponding to the average values ​​of the unreversed indicators m1, ..., m N , Based on the average value of the fall index and N real sensing values ​​p1, ....p N Calculate the uninverted distance l: The integer y in the above equation takes values ​​from 1 to N.

Images