Sensor

Abstract

The sensor of the present invention includes: a complex transfer function calculation unit (30) that calculates a complex transfer function based on a received signal obtained by a receiving antenna element receiving a signal transmitted from a transmitter (10) to a predetermined space during a predetermined period; a biological component extraction unit (40) that extracts biological information as a component corresponding to a living body that may exist in the predetermined space; a correlation matrix calculation unit (50) that calculates a correlation matrix based on the biological information extracted by the biological component extraction unit (40); a first number of people information calculation unit (60) that calculates first number of people information; a MUSIC spectrum calculation unit (70) that estimates candidates for the positions of one or more living bodies by a predetermined position estimation method using the correlation matrix and outputs a likelihood spectrum indicating that a living body exists at the position; and a second number of people information calculation unit (80) that estimates second number of people information or a position as the number of living bodies with higher confidence based on first position information that can include a plurality of position candidates by a predetermined method.

Description

Technical Field

[0001] The present invention relates to a sensor, and particularly to a sensor for estimating the number of living organisms using a wireless signal. Background Art

[0002] Techniques for detecting a detection target using a signal transmitted wirelessly have been developed (for example, see Patent Document 1).

[0003] Patent Document 1 discloses a technique in which a characteristic value including a component of Doppler shift is analyzed using Fourier transform on a signal received wirelessly, so that the number and position of a person as a detection target can be known.

[0004] Prior Art Documents

[0005] Patent Documents

[0006] Patent Document 1: Japanese Unexamined Patent Application Publication No. 2015-117972

[0007] Patent Document 2: Japanese Unexamined Patent Application Publication No. 2014-228291

[0008] Patent Document 3: Japanese Patent No. 5047002

[0009] Patent Document 4: Japanese Patent No. 5025170 Summary of the Invention

[0010] Problems to be Solved by the Invention

[0011] However, in the technique disclosed in Patent Document 1, since the detectable distance becomes short in a situation where the Doppler effect is very weak, such as a stationary living organism, as the detection target, there is a problem that the range in which the number of people can be estimated and the position of the living organism can be estimated becomes narrow.

[0012] The present invention has been made in view of the above circumstances, and an object thereof is to provide a sensor capable of estimating the position and number of living organisms in a wider range using a wireless signal.

[0013] Means for Solving the Problems

[0014] To achieve the above object, a sensor according to one aspect of the present invention includes: a complex transfer function calculation unit that calculates a complex transfer function based on received signals obtained by M (M is a natural number of 2 or more) receiving antenna elements receiving signals transmitted from a transmitter having N (N is a natural number of 2 or more) transmitting antenna elements to a predetermined space during a predetermined period; a biological component extraction unit that extracts biological information as a component corresponding to a living body that may exist in the predetermined space; a correlation matrix calculation unit that calculates a correlation matrix based on the biological information extracted by the biological component extraction unit; a first number of people information calculation unit that calculates first number of people information as an assumed value of the number of people existing in the predetermined space; a likelihood spectrum calculation unit that estimates candidates for the position of the living body using the correlation matrix by a predetermined position estimation method and outputs a likelihood spectrum indicating that the living body exists at the position; and a second number of people information calculation unit that estimates second number of people information or a position as the number of living bodies with higher confidence based on first position information that can include a plurality of position candidates by a predetermined method.

[0015] Advantages of the Invention

[0016] With the sensor according to the present invention, it is possible to more accurately estimate the number of living bodies using wireless signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a block diagram showing the structure of the sensor of Embodiment 1.

[0018] Figure 2 It is a conceptual diagram of an area where the sensor of Embodiment 1 is installed.

[0019] Figure 3 It is a block diagram showing the structure of the second number of people information calculation unit using the ratio method.

[0020] Figure 4 It is a conceptual diagram showing the operation of the peak search unit of Embodiment 1.

[0021] Figure 5 It is a conceptual diagram showing the operation of the ratio test unit of Embodiment 1.

[0022] Figure 6 It is a flowchart showing the processing of the sensor of Embodiment 1.

[0023] Figure 7 It is a flowchart showing the second number of people information calculation process of the sensor of Embodiment 1.

[0024] Figure 8 It is a block diagram showing the structure of the second number of people information calculation unit of Embodiment 2.

[0025] Figure 9It is a conceptual diagram showing the operation of the block detection unit according to Embodiment 2.

[0026] Figure 10 It is a block diagram showing the configuration of the second number-of-people information calculation unit according to Embodiment 3. Detailed Embodiment

[0027] (Understanding underlying the present invention)

[0028] Techniques for detecting a detection target using signals transmitted wirelessly have been developed (for example, refer to Patent Documents 1 to 4).

[0029] For example, in Patent Document 1, a technique is disclosed for estimating the number and position of a person as a detection target by analyzing eigenvalue components including Doppler frequency shift using Fourier transform. Specifically, the received signal is subjected to Fourier transform, an autocorrelation matrix is obtained for a waveform that extracts specific frequency components, and eigenvalues are obtained by performing eigenvalue decomposition on the autocorrelation matrix. Usually, an eigenvalue and an eigenvector represent one propagation path of radio waves from a transmitting antenna to a receiving antenna, that is, a path. However, in the technique of Patent Document 1, since components that do not contain biological information are removed, only the paths corresponding to the signals reflected by the organism and the paths corresponding to their secondary reflections and noise are presented in the eigenvalues and eigenvectors. Here, since the value of the eigenvalue corresponding to noise is smaller than the value of the eigenvalue corresponding to the organism, by enumerating the number of eigenvalues larger than a specified threshold, the number of organisms can be estimated.

[0030] However, in the technique disclosed in Patent Document 1, when the target organism is at a distance or the number of organisms is large, there is a problem that the difference between the eigenvalue corresponding to the organism and the eigenvalue corresponding to the noise decreases and the accuracy of number-of-people estimation deteriorates. This is because, in a situation where the Doppler effect is very weak, it is difficult to detect weak signals that cause Doppler frequency shift due to the influence of internal noise of the receiver, interference waves coming from outside the detection target, and the presence of objects that generate Doppler frequency shift outside the detection target. In addition, since the organism as a measurement target has a certain size and the components of the organism are distributed across multiple eigenvalues, when the number of organisms increases, the separation of the eigenvalues of the organism cannot be completed completely, and it becomes difficult to estimate the number of people.

[0031] In Patent Document 2, a technique for estimating the position of an object using a direction estimation algorithm such as the MUSIC (MUltiple SIgnal Classification) method is disclosed. Specifically, a receiving station that receives a signal transmitted from a transmitting station performs a Fourier transform on the received signal, obtains an autocorrelation matrix for a waveform from which a specific frequency component has been extracted, and applies a direction estimation algorithm such as the MUSIC method. Thereby, it is possible to perform direction estimation of a living body with relatively high accuracy. However, since the MUSIC method used in Patent Document 2 requires the number of living bodies to be detected in advance, it can be seen that the technique in Patent Document 2 requires human number estimation in advance.

[0032] In addition, for example, in Patent Document 3, a technique for estimating the number of incoming waves, that is, the number of transmitters such as mobile phones, based on the correlation between the eigenvector of the received signal received by multiple antennas and the steering vector of the range where the radio wave may arrive is disclosed.

[0033] In addition, for example, in Patent Document 4, a technique is disclosed in which various numbers of incoming waves are assumed for the received signal received by multiple antennas, an evaluation function using the steering vector is calculated for each of them, and the number of incoming waves with the maximum evaluation function is estimated as the true number of incoming waves.

[0034] However, the techniques disclosed in Patent Documents 3 to 4 are techniques for estimating the number of transmitters that emit radio waves, and it can be seen that the number of living bodies cannot be estimated.

[0035] Therefore, in view of these, the inventors have found a sensor or the like that can use wireless signals to more accurately estimate a larger number of living bodies without having a special device such as a transmitter held by the living body to be targeted, and have achieved the present invention.

[0036] That is, a sensor according to one aspect of the present invention includes: a complex transfer function calculation unit that calculates a complex transfer function based on received signals received by M (M is a natural number of 2 or more) receiving antenna elements from a transmitter having N (N is a natural number of 2 or more) transmitting antenna elements to a specified space during a specified period; a biological component extraction unit that extracts biological information as a component corresponding to a living body that may exist in the specified space; a correlation matrix calculation unit that calculates a correlation matrix based on the biological information extracted by the biological component extraction unit; a first number of people information calculation unit that calculates first number of people information as an assumed value of the number of people existing in the specified space; a likelihood spectrum calculation unit that estimates candidates for the position of the living body using the correlation matrix by a specified position estimation method and outputs a likelihood spectrum of the living body existing at that position; and a second number of people information calculation unit that estimates second number of people information or a position as the number of living bodies with higher confidence based on first position information that can include multiple position candidates by a specified method.

[0037] With this configuration, even when a biological component is superimposed on eigenvalues that should originally be separated as noise, it is possible to improve the accuracy of estimating the number of people by determining whether the peak of the spectrum is a true peak based on the shape of the likelihood spectrum.

[0038] First, in cases where the signal-to-noise ratio is small, such as when the living body is far away, it is difficult to distinguish between eigenvalues corresponding to the living body and eigenvalues corresponding to noise using a threshold value method based on eigenvalues. In addition, when the living body being measured has a certain size and the number of living bodies increases because the components of the living body are distributed across multiple eigenvalues, it is not possible to completely separate the eigenvalues of the living body, making it difficult to estimate the number of people.

[0039] Therefore, in the present invention, first number of people information, which is the number of people that can be measured based on the environment where the sensor is installed and the number of antenna elements of the sensor, is assumed, and a likelihood spectrum is calculated. Thus, there may be a case where the actual number of people existing in the detection range is different from the first number of people information used in the calculation of the likelihood spectrum.

[0040] A representative likelihood spectrum is a MUSIC spectrum based on the MUSIC method, and the MUSIC spectrum will be described hereinafter as a representative.

[0041] In the MUSIC method, when the first number of people is less than the number of people actually present, the number of peaks in the MUSIC spectrum is less than the number of people actually present, so omissions occur. In addition, when the first number of people is more than the number of people actually present, a peak called a virtual image appears in the MUSIC spectrum at a position where there is no actual living body, but the inventors have found that the virtual image has characteristics such as a low peak height and smoothness. Therefore, the inventors have found that for the MUSIC spectrum, by distinguishing the virtual image by using a threshold-based determination of the peak, a determination based on the ratio of each peak, and a determination based on machine learning, and counting the number of real peaks, it is possible to perform a high-precision estimation of the number of people.

[0042] Furthermore, for example, the first number of persons information calculation unit may be configured to output a fixed value obtained by a predetermined method based on the number N of the transmitting antenna elements and the number M of the receiving antenna elements as the first number of persons information.

[0043] With this configuration, it is possible to set appropriate first number of people information for the number of antenna elements constituting the sensor, thereby improving the accuracy of the obtained likelihood spectrum.

[0044] In addition, the present invention can be realized not only as a device but also as an integrated circuit having a processing mechanism of such a device, or as a method with the processing mechanism constituting the device as steps, or as a program for causing a computer to execute these steps, or as information, data or signals representing the program. Furthermore, these programs, information, data and signals can also be distributed via a recording medium such as a CD-ROM or a communication medium such as the Internet.

[0045] Hereinafter, the embodiments of the present invention will be described using the accompanying drawings. In addition, the embodiments described below all represent a preferred specific example of the present invention. The numerical values, shapes, materials, constituent elements, configuration positions of constituent elements and connection forms, steps, order of steps, etc. shown in the following embodiments are examples and are not intended to limit the present invention. In addition, regarding the constituent elements of the following embodiments that are not recorded in the independent claims representing the highest concept of the present invention, they are described as arbitrary constituent elements that constitute a more preferred form. In addition, in this specification and the accompanying drawings, regarding constituent elements that have substantially the same functional structure, repeated descriptions are omitted by assigning the same reference numerals.

[0046] (Implementation Method 1)

[0047] Below, while referring to the attached Figure 1 The method of estimating the number of people by the sensor 1 according to the first embodiment will be described.

[0048] [Structure of sensor 1]

[0049] Figure 1 It is a block diagram showing the structure of the sensor 1 of Embodiment 1. Figure 2 It is a conceptual diagram of the area where the sensor 1 of Embodiment 1 is provided.

[0050] Figure 1 The shown sensor 1 includes a transmitter 10, a receiver 20, a complex transfer function calculation unit 30, a biological component extraction unit 40, a correlation matrix calculation unit 50, a first number information calculation unit 60, a MUSIC spectrum calculation unit 70, and a second number information calculation unit 80.

[0051] [Transmitter 10]

[0052] The transmitter 10 has a transmitting antenna. Specifically, the transmitter 10 is composed of a transmitting unit 11 and a transmitting antenna unit 12 as Figure 1 shown. The transmitting antenna unit 12 is composed of an array antenna of M T elements. For example, it is a 4-element patch array antenna with an array element antenna spacing of half a wavelength, etc.

[0053] The transmitting unit 11 generates a high-frequency signal used to estimate the presence, position, and number of the living body 200. For example, the transmitting unit 11 generates a CW (Continuous Wave) of 2.4 GHz and transmits the generated CW as a transmission wave from the transmitting antenna unit 12. In addition, the transmitted signal is not limited to CW and may also be a modulated signal.

[0054] [Receiver 20]

[0055] The receiver 20 includes a receiving antenna unit 21 and a receiving unit 22.

[0056] The receiving antenna unit 21 is composed of an array antenna of M R elements. For example, it is a 4-element patch array antenna with an array element antenna spacing of half a wavelength, etc. The receiving antenna unit 21 receives a high-frequency signal with the array antenna.

[0057] The receiving unit 22 converts the high-frequency signal received by the receiving antenna unit 21 into a low-frequency signal that can be processed for signal processing, for example, using a downconverter. In addition, when the transmitter 10 transmits a modulated signal, the receiving unit 22 also demodulates the received modulated signal. The receiving unit 22 transmits the converted low-frequency signal to the complex transfer function calculation unit 30.

[0058] In addition, the frequency used as an example in this embodiment is 2.4 GHz, but frequencies such as 5 GHz or the millimeter wave band can also be used.

[0059] [Complex transfer function calculation unit 30]

[0060] The complex transfer function calculation unit 30 calculates a complex transfer function representing the propagation characteristics between the array antenna of the receiving antenna unit 21 and the transmitting antenna unit 12 of the transmitter 10 based on the signals observed by the array antenna of the receiving antenna unit 21. More specifically, the complex transfer function calculation unit 30 calculates, based on the low-frequency signals passed through the receiving unit 22, a complex transfer function representing the propagation characteristics between the M T transmitting antenna elements of the transmitting antenna unit 12 and the M R receiving antenna elements of the receiving antenna unit. In addition, in the complex transfer function calculated by the complex transfer function calculation unit 30, there is a case where a part of the transmitted wave transmitted from the transmitting antenna unit 12 includes a reflected wave or a scattered wave that is a signal reflected or scattered by the living body 200. In addition, in the complex transfer function calculated by the complex transfer function calculation unit 30, it includes a direct wave from the transmitting antenna unit 12 and a reflected wave from a fixed object, etc., which are reflected waves that do not pass through the living body 200. In addition, the amplitude and phase of the signal reflected or scattered by the living body 200, that is, the reflected wave and the scattered wave from the living body 200, always change due to the biological activities such as the respiration and heartbeat of the living body 200.

[0061] Hereinafter, it is assumed that the complex transfer function calculated by the complex transfer function calculation unit 30 includes a reflected wave and a scattered wave that are signals reflected or scattered by the living body 200 for explanation.

[0062] In addition, in the Figure 1 structural diagram, the transmitter 10 and the receiver 20 are shown adjacent to each other, but in reality, they can be arranged at a distance from each other as shown in Figure 2 . In addition, the transmitting antenna and the receiving antenna can also be used interchangeably. In addition, it can also be shared with the hardware of wireless devices such as a Wi-Fi router or a handset.

[0063] [Biological component extraction unit 40]

[0064] The biological component extraction unit 40 extracts biological components from the signals observed by the receiving array antenna of the receiving antenna unit 21, which are signal components that are transmitted from the transmitting antenna unit 12 and reflected or scattered by one or more organisms 200. More specifically, the biological component extraction unit 40 records the complex transfer function calculated by the complex transfer function calculation unit 30 in the order in which the signals are observed, that is, in time series. And the biological component extraction unit 40 extracts the variation components caused by the influence of the organism 200 from the variations of the complex transfer function recorded in time series. Hereinafter, the variation components of the complex transfer function caused by the influence of the organism 200 are referred to as biological components. Here, as a method for extracting biological components, for example, there is a method of transforming into the frequency domain by Fourier transform or the like and then extracting only biological components, or a method of extracting by calculating the difference between the complex transfer functions at two different times. By these methods, the components directly affecting the reflected wave via the fixture are removed, and only the biological components via the organism 200 are left. For example, using the complex transfer function for 5 seconds, the components from 0.3 Hz to 3 Hz are extracted, and the respiratory components that also exist when the organism is at rest are extracted.

[0065] In addition, in the present embodiment, as an example, the components from 0.3 Hz to 3 Hz are extracted. However, when it is desired to extract slower or faster movements, of course, it is only necessary to change the extraction frequency components accordingly.

[0066] In addition, in the present embodiment, since there are M T transmitting antenna elements constituting the transmitting array antenna and M R receiving antenna elements constituting the receiving array antenna, that is, there are multiple, the biological components via the organism 200 of the complex transfer function corresponding to the transmitting and receiving array antennas also become multiple. Hereinafter, they are collectively referred to as the biological component channel matrix F(f) of M rows and N columns and are expressed as in (Equation 1).

[0067] [Equation 1]

[0068]

[0069] In addition, each element F ij of the biological component complex transfer function matrix, that is, the biological component channel matrix F(f), is an element obtained by extracting the variation components from each element h ij of the complex transfer function. In addition, the biological component complex transfer function matrix, that is, the biological component channel matrix F(f), is a function of frequency or a differential period similar thereto and contains information corresponding to multiple frequencies.

[0070] [Correlation matrix calculation unit 50]

[0071] The correlation matrix calculation unit 50 rearranges the elements of the biological component channel matrix composed of M rows and N columns calculated by the biological component extraction unit 40, and transforms it into a biological component channel vector F of M×N rows and 1 column vec (f). As an arrangement method, there is a method such as (Equation 2), but as long as it is an operation of rearranging the matrix, the order is not limited.

[0072] [Equation 2]

[0073] F vec (f) = vec[F(f)] = [F 11 (f)…F M1 (f)F 12 (f)…F M2 (f)…F 1N (f)…F MN (f)] T (Equation 2)

[0074] Then, the correlation matrix calculation unit 50 calculates the correlation matrix based on the biological component channel vector F vec (f). More specifically, the correlation matrix calculation unit 50 calculates the correlation matrix R of the biological component channel vector F vec (f) composed of multiple varying components brought by the organism 200 according to (Equation 3).

[0075] [Equation 3]

[0076] R = E[F vec (f)F vec (f) H (Equation 3)

[0077] E[] in (Equation 3) represents the averaging operation, and the operator H represents the complex conjugate transpose. Here, by averaging the biological component channel vector F vec (f) containing multiple frequency components in the frequency direction, sensing that can simultaneously use the information contained in each frequency can be performed.

[0078] [First number information calculation unit 60]

[0079] The first number information calculation unit 60 calculates the first number information used by the MUSIC spectrum calculation unit 70 according to a prescribed method. Here, the first number information is 1 or more and less than M R ×M TAn integer. For the method used in calculating the first number of people information, there are the following methods. For example, it can be a method using a fixed value obtained from the number of transmitting antennas and receiving antennas, or it can be a method using, for example, the average value of the number of transmitting antennas and the number of receiving antennas. In addition, when the maximum number of people can be specified based on the information of the place where the sensor 1 is installed, this value can also be used as the first number of people information. In addition, the first number of people information does not need to be a fixed value. For example, the number of people estimated by the past sensor 1 can also be stored, and a value about 1 to 2 more than this number can be used as the first number of people information.

[0080] Originally, in order to calculate the MUSIC spectrum, the arrival wave number is required, and in this embodiment, the number of people is required to be known. However, by calculating the first number of people information as an assumed number of people by the first number of people information calculation unit 60, the MUSIC spectrum can be calculated.

[0081] [MUSIC spectrum calculation unit 70]

[0082] If the correlation matrix calculated by the correlation matrix calculation unit 50 is subjected to eigenvalue decomposition, then

[0083] [Equation 4]

[0084] R = UAU H

[0085] [Equation 5]

[0086]

[0087] [Equation 6]

[0088]

[0089] Here,

[0090] [Equation 7]

[0091]

[0092] is the eigenvector with the number of elements being M R and

[0093] [Equation 8]

[0094]

[0095] is the eigenvalue corresponding to the eigenvector, and is set as

[0096] [Equation 9]

[0097] in this order. In addition, L is the first number of people information calculated by the first number of people information calculation unit 60.

[0098] In addition, the steering vector (direction vector) of the transmitting array antenna is defined as

[0099] [Equation 10]

[0100]

[0101] The steering vector (direction vector) of the receiving array antenna is defined as

[0102] [Equation 11]

[0103]

[0104] Here, k is the wave number. Furthermore, multiplying these steering vectors is defined as the steering vector considering the angular information of both the transmitting and receiving array antennas

[0105] [Equation 12]

[0106]

[0107] Apply the MUSIC method to it.

[0108] That is, the MUSIC spectrum calculation unit 70 calculates the evaluation function P represented by the following (Equation 4) based on the MUSIC method and using the multiplied steering vector music spectrum of (θ).

[0109] [Equation 13]

[0110]

[0111] In addition, the MUSIC spectrum can also be replaced by a spectrum based on the Beamformer method or the Capon method. However, it should be noted that they have lower accuracy compared to the MUSIC method and cannot perform high-precision estimation alone.

[0112] [Second number information calculation unit 80]

[0113] The second number information calculation unit 80 calculates second number information as the number of living organisms existing in the target area based on the MUSIC spectrum calculated by the MUSIC spectrum calculation unit 70.

[0114] Originally, a peak corresponding to the correct number of people appears in the MUSIC spectrum calculated based on the correct number of people. However, in the present embodiment, since the MUSIC spectrum is calculated using the first number information that is greater than the correct number of people, a false peak that appears as a peak although there is actually no living body occurs. Therefore, in the second number information calculation unit 80, a peak that is not a false peak among the peaks that appear in the MUSIC spectrum is discriminated. For the calculation of the second number information, there are, for example, a method of using the ratio method for the peak of the spectrum, a method of counting the number of consecutive blocks in a region above a specified threshold in the MUSIC spectrum, or a method of using machine learning such as a convolutional neural network by treating the MUSIC spectrum as an image. In the present embodiment, as an example, a method for calculating the second number information using the ratio method will be described.

[0115] Figure 3 FIG. is a block diagram of the second number information calculation unit 80 that calculates the second number information using the ratio method. The second number information calculation unit 80 includes a peak search unit 81, a false peak determination unit 82, a peak sorting unit 83, and a ratio test unit 84.

[0116] <Peak search unit 81>

[0117] The peak search unit 81 searches for peaks that take the maximum value in the MUSIC spectrum. The set of peaks found here is set as the first peak set. In addition, in order to exclude small peaks caused by noise, it is preferable to limit only to peaks that are the maximum value within a specified range x.

[0118] Figure 4 FIG. is a conceptual diagram showing the operation of the peak search unit 81 of Embodiment 1, and the processing of the peak search unit 81 is represented by a one-dimensional MUSIC spectrum 1000.

[0119] In Figure 4 there are four peaks, 1001 - A, 1001 - B, 1001 - C, and 1001 - D. The peaks that are the maximum value within a range of 0.5 m from the peaks, represented by 1002 - A, 1002 - B, 1002 - C, and 1002 - D respectively, are three peaks, 1001 - A, 1001 - B, and 1001 - D. Therefore, the peak search unit 81 extracts three peaks, 1001 - A, 1001 - B, and 1001 - D, from the MUSIC spectrum 1000 as the first peak set.

[0120] <False peak determination unit 82>

[0121] The false peak determination unit 82 calculates the y% value of the values included in a specified x range around the peak included in the first peak set, and extracts those for which the difference or ratio between the peak and the y% value is equal to or greater than a specified threshold z as the second peak set. Thereby, it is possible to exclude those with smooth peaks from the peaks included in the first peak set. For example, when the specified range is a radius of 0.5 m, y is 70%, and z is 0.4 dB, for each peak included in the first peak set, only the peaks greater than 0.4 dB above the 70% value of the values included in the surrounding radius of 0.5 m are extracted.

[0122] <Peak sorting unit 83>

[0123] The peak sorting unit 83 sorts the peaks of each second peak set in descending order. Additionally, the peak sorting unit 83 may append, for the second peak set, a value w smaller than the smallest of the peaks included in the second peak set as a virtual peak. For example, when w is set to 3.4 dB and the smallest peak is -3 dB relative to the largest peak, the appended virtual peak is -6.4 dB relative to the largest peak.

[0124] <Ratio test unit 84>

[0125] The ratio test unit 84 estimates the second person information by calculating the ratio between adjacent peaks of the second peak set sorted by the peak sorting unit 83. More specifically, it calculates the ratio or difference between the i-th peak and the (i + 1)-th peak of the second peak set sorted in descending order, and outputs the i for which this ratio or difference is the largest as the second person information. Here, i is an integer greater than or equal to 1 and less than or equal to the number of elements in the second peak set.

[0126] Figure 5 is a conceptual diagram showing the operation of the ratio test unit 84 of Embodiment 1.

[0127] In Figure 5 it, the elements 1101 - A, 1101 - B, 1101 - C, 1102 of the second peak set are recorded sorted in descending order according to the peaks. Additionally, the peak 1102 is a virtual peak appended by the peak sorting unit 83.

[0128] The peak sorting unit 83 calculates the differences 1103 - A, 1103 - B, 1103 - C between adjacent peaks of the second peak set, and obtains the combination of peaks for which the difference is the largest. In Figure 5 the example shown, since 1103 - B is the largest and the difference between the second peak 1101 - B and the third peak 1101 - C is the largest, the calculated second person information is 2.

[0129] In addition, in this embodiment, since the purpose is to estimate the number of people, the second number information is used as the output of the sensor. However, the MUSIC spectrum can also be used to estimate the position of the living body, and the position information of the living body can be used as the output of the sensor.

[0130] In addition, in this embodiment, an example in which both the transmitting antenna and the receiving antenna have a multiple-input multiple-output (MIMO) structure is described. However, a structure in which either the transmitting or receiving side has a single antenna can also be used. In this case, the MUSIC spectrum output by the MUSIC spectrum calculation unit is one-dimensional. However, in this case, the estimation of the second number information based on peak search can also be performed in the same manner as in the two-dimensional case.

[0131] In addition, it is also possible to detect the presence or absence, that is, the case of 0 people, and make a determination based on the magnitude of the largest eigenvalue and the power of the fluctuation component of the complex transfer function and the correlation value when there is no one. The MUSIC spectrum calculation by the MUSIC spectrum calculation unit 70 is performed only when there are people. By doing so, the calculation required for MUSIC spectrum calculation can be omitted in the case of no people.

[0132] [Operation of Sensor 1]

[0133] The process of estimating the number of living bodies of the sensor 1 configured as described above will be described.

[0134] Figure 6 It is a flowchart showing the process of the sensor 1 of Embodiment 1. Figure 7 It is a flowchart showing the second number information calculation process of the sensor of Embodiment 1.

[0135] First, as shown in Figure 6 , the sensor 1 observes the received signal (S10) in the receiver 20 for a specified period.

[0136] Next, the sensor 1 calculates the complex transfer function based on the received signal (S20).

[0137] Next, the sensor 1 records each calculated complex transfer function in time series, and calculates the living body component channel matrix by extracting the fluctuation component caused by the influence of the living body from the recorded time series of complex transfer functions (S30).

[0138] Next, the sensor 1 calculates the correlation matrix of the extracted living body component channel matrix (S40). Next, the sensor 1 calculates the first number information by a specified method (S50). The specified method described here can be either a fixed value based on the number of transmitting antenna elements and receiving antenna elements provided in the sensor 1 or a variable value based on the past number estimation results.

[0139] Next, based on the first number of people information calculated in S50 and the correlation matrix calculated in S40, sensor 1 calculates the MUSIC spectrum (S60).

[0140] Finally, sensor 1 calculates the second number of people information based on the MUSIC spectrum calculated in S60 and outputs it as the number of organisms (S70). The process of S70 is performed by, for example, using a machine learning method such as a neural network by treating the MUSIC spectrum as an image, a method of counting the number of consecutive blocks of regions with values above a specified value in the MUSIC spectrum, or a method of using a ratio method for the peaks of the spectrum, etc.

[0141] Hereinafter, assuming that the process of S70 is performed using the ratio method, the following Figure 7 flowchart is used for explanation.

[0142] In S70, first, as shown in Figure 7 , sensor 1 extracts the peak that is the maximum value within a specified range among the peaks of the MUSIC spectrum as the first peak set (S71).

[0143] Next, sensor 1 calculates the y% value of the values included in a specified range around each peak included in the first peak set (S72).

[0144] Next, sensor 1 extracts the peaks whose difference between the y% value and the peak is above a specified threshold as the second peak set (S73).

[0145] Next, sensor 1 sorts the peaks included in the second peak set in descending order (S74).

[0146] Finally, sensor 1 calculates the ratio or difference between the i-th peak and the (i + 1)-th peak of the second peak set, and outputs the i for which the ratio or difference is the largest as the second number of people information. Here, i is an integer greater than or equal to 1 and less than or equal to the number of elements in the second peak set.

[0147] [Effects, etc.]

[0148] The sensor 1 according to the present embodiment can accurately estimate the number of organisms present using wireless signals. In addition, the sensor 1 according to the present embodiment can accurately estimate the position of organisms even when the number of organisms is large and noise and organism components cannot be sufficiently separated by eigenvalue decomposition. More specifically, in the existing method for estimating the number of people using eigenvalues, it is based on the premise that the eigenvalues corresponding to noise and the eigenvalues corresponding to organisms can be clearly distinguished. However, in reality, there is also a slight overlay of organism components in the noise component, resulting in a decrease in the accuracy of the existing method for estimating the number of organisms. In the present invention, the calculation of the spectrum based on the MUSIC method is performed using first person number information that may be larger than the actual number of people. Even if there are errors in the peaks of the spectrum, by extracting only the true peaks based on features, the number of people can be estimated even when the eigenvalues of noise and the eigenvalues of organisms cannot be sufficiently separated.

[0149] (Modification of Embodiment 1)

[0150] In addition, as a modification of Embodiment 1, the complex transfer function calculation unit 30 may also directly transfer the calculated complex transfer function to the second person number information calculation unit 80, and the second person number information calculation unit 80 calculates the second person number information by a machine learning method such as the k-nearest neighbor method using the complex transfer function as input.

[0151] By making this modification, the calculation of the second person number information can be performed without being affected by the region where the MUSIC spectrum is calculated.

[0152] (Embodiment 2)

[0153] In Embodiment 1, the method of calculating the second person number information by the second person number information calculation unit 80 using the ratio method was described. In Embodiment 2, the method of the second person number information calculation unit 2080 using the method of counting the number of consecutive blocks in the region above a specified threshold in the MUSIC spectrum is described using the block diagram of Figure 8 and the conceptual diagram of Figure 9 . In addition, since the structures other than the second person number information calculation unit 2080 are the same as those in Embodiment 1, the description thereof is omitted here.

[0154] Figure 8 is a block diagram showing the structure of the second person number information calculation unit 2080 of Embodiment 2. Figure 9 is a conceptual diagram showing the operation of the block detection unit 2082 of Embodiment 2.

[0155] The second person number information calculation unit 2080, as shown in Figure 8 , includes a threshold setting unit 2081 and a block detection unit 2082.

[0156] The threshold setting unit 2081 sets a threshold 2101 that is v dB less than the maximum value of the MUSIC spectrum 2100 for the MUSIC spectrum 2100. Additionally, for v and the threshold 2101, either a preset fixed value can be used, or v and the threshold 2101 can be varied in various ways beforehand to evaluate the accuracy of the number of people estimation, and the threshold 2101 with the highest accuracy is used as the optimal value. For example, in the case of using an unmodulated continuous wave of 2.47125 GHz and sensing a 4 m four-sided room with a 4-element patch array antenna with an element spacing of half a wavelength, v is set to 3.9 dB.

[0157] The block detection unit 2082 detects the regions where the MUSIC spectrum 2100 is above the threshold 2101 and counts the number. In Figure 9 the example shown, two of the regions 2102 - A and 2102 - B are blocks above the threshold 2101, and the second person number information is calculated as 2.

[0158] [Effects, etc.]

[0159] Compared with the sensor of Embodiment 1, the sensor according to Embodiment 2 can reduce the calculation amount in the second person number information calculation unit, can reduce the capabilities of the processing device required for real-time processing, and can achieve the number of people estimation at low cost.

[0160] (Embodiment 3)

[0161] In Embodiment 1, the method of calculating the second person number information by the second person number information calculation unit 80 using the ratio method was described. In Embodiment 3, using Figure 10 the block diagram, the case where the second person number information calculation unit 3080 calculates the second person number information using a convolutional neural network is described. Additionally, since the structures other than the second person number information calculation unit 3080 are the same as those in Embodiment 1, the description thereof is omitted here.

[0162] Figure 10 is a block diagram showing the structure of the second person number information calculation unit of Embodiment 3.

[0163] The second person number information calculation unit 3080 is as Figure 10 shown, including a teacher data production unit 3081, a learning unit 3082, a network storage unit 3083, an image transformation unit 3084, and a determination unit 3085.

[0164] The teacher data production unit 3081, the learning unit 3082, and the network storage unit 3083 perform the pre-learning of the network, and the image transformation unit 3084 and the determination unit 3085 use the pre-learned network to calculate the second person number information for the test data.

[0165] The teacher data production department 3081 acquires MUSIC spectra in advance when multiple known numbers of people are present, and saves them as images of teacher data. Here, multiple or more teacher data images are prepared for all conceivable numbers of people. For example, when the upper limit of the number of people present in the measurement area is 3, multiple, e.g., 100 or more, teacher data images are prepared for 0 people, 1 person, 2 people, and 3 people respectively.

[0166] The learning department 3082 uses the teacher data images as input to perform learning of a convolutional neural network. Additionally, here, a method such as transfer learning that makes the learning of the neural network more efficient can also be used.

[0167] The network storage department 3083 stores the convolutional neural network learned by the learning department 3082 in a memory on a computer or a recording medium such as a CD-ROM, or stores it in an external server via a network.

[0168] The image transformation department 3084 transforms the MUSIC spectrum output by the MUSIC spectrum calculation unit 70 into a form that can be processed by the convolutional neural network. For example, a bitmap image in which each pixel corresponds to the value of the MUSIC spectrum can be processed by the convolutional neural network.

[0169] The determination department 3085 performs determination using the convolutional neural network of the network stored in the network storage department 3083, and outputs the output of the convolutional neural network as second person number information.

[0170] Above, a sensor related to one technical solution of the present invention has been described based on the embodiments, but the present invention is not limited to these embodiments. As long as it does not deviate from the gist of the present invention, forms obtained by various modifications that those skilled in the art can think of to these embodiments, or forms constructed by combining the constituent elements of different embodiments are also included in the scope of the present invention.

[0171] In addition, the present invention has such characteristic constituent elements, and can be implemented not only as a sensor, but also as an estimation method etc. with the characteristic constituent elements included in the sensor as steps. In addition, it can also be implemented as a computer program that causes a computer to execute each characteristic step included in such a method. And of course, such a computer program can be circulated via a non-volatile recording medium such as a CD-ROM that can be read by a computer or a communication network such as the Internet.

[0172] [Effects, etc.]

[0173] By performing machine learning based on a convolutional neural network using the sensor of Embodiment 3, it is possible to automatically perform various parameter adjustments such as thresholds that need to be changed separately according to the environment where the sensor is set. In addition, by updating the learned network at any time, further improvement in the accuracy of estimating the number of people can be expected.

[0174] Industrial applicability

[0175] The present invention can be used in a measuring device for measuring the number and position of living organisms, a home appliance device for performing control corresponding to the number and position of living organisms, a monitoring device for detecting the intrusion of living organisms, and the like.

[0176] Reference numeral description

[0177] 1 Sensor

[0178] 10 Transmitter

[0179] 11 Transmission unit

[0180] 12 Transmission antenna unit

[0181] 20 Receiver

[0182] 21 Reception antenna unit

[0183] 22 Reception unit

[0184] 30 Complex transfer function calculation unit

[0185] 40 Biological component extraction unit

[0186] 50 Correlation matrix calculation unit

[0187] 60 First number-of-people information calculation unit

[0188] 70 MUSIC spectrum calculation unit

[0189] 80, 2080, 3080 Second number-of-people information calculation unit

[0190] 81 Peak exploration unit

[0191] 82 False peak determination unit

[0192] 83 Peak sorting unit

[0193] 84 Ratio test unit

[0194] 200 Organism

[0195] 1000, 2100 MUSIC spectrum

[0196] 1001 - A, 1001 - B, 1001 - C, 1001 - D, 1102 Peak

[0197] Regions defined around the peaks of 1002 - A, 1002 - B, 1002 - C, and 1002 - D

[0198] Peaks included in the second peak set of 1101 - A, 1101 - B, and 1101 - C

[0199] Differences between adjacent peaks of 1103 - A, 1103 - B, and 1103 - C

[0200] 2081 Threshold setting unit

[0201] 2082 Block detection unit

[0202] 2101 Threshold

[0203] Regions where the MUSIC spectrum of 2102 - A and 2102 - B is above the threshold

[0204] 3081 Teacher data production unit

[0205] 3082 Learning unit

[0206] 3083 Network storage unit

[0207] 3084 Image transformation unit

[0208] 3085 Judgment unit

Claims

1. A sensor for estimating the number of organisms, comprising: A complex transfer function calculation unit that calculates a complex transfer function based on received signals obtained by M receiving antenna elements receiving signals transmitted from a transmitter with N transmitting antenna elements to a specified space during a specified period, where M and N are natural numbers greater than or equal to 2; An organism component extraction unit that extracts organism information as a component corresponding to an organism that may exist in the specified space; A correlation matrix calculation unit calculates a correlation matrix based on the biological information extracted by the above-described biological component extraction unit; A first number of people information calculation unit calculates first number of people information that is an assumed value of the number of people present in the above-described specified space; A likelihood spectrum calculation unit estimates candidates for the positions of the above-described organisms using the above-described correlation matrix and the above-described first number of people information, and outputs a likelihood spectrum indicating the likelihood that the organisms are present at those positions, by a specified position estimation method; and A second number of people information calculation unit estimates second number of people information that is the number of organisms with a higher confidence level than the above-described first number of people information, based on the above-described likelihood spectrum and according to first position information that can include multiple position candidates, by a specified method.

2. The sensor according to claim 1, The first number information calculation unit outputs a fixed value obtained by a specified method based on the number N of the transmitting antenna elements and the number M of the receiving antenna elements as the first number information.

3. The sensor according to claim 1, The first number information calculation unit outputs the first number information based on the maximum number of people that can exist in the specified space where the transmitter transmits signals.

4. The sensor according to claim 1, It further comprises a storage unit for storing the number information estimated by the second number information calculation unit; The first number information calculation unit calculates the first number information by a specified method based on the past number information stored in the storage unit.

5. The sensor according to any one of claims 1 to 4, The second number information calculation unit estimates the number of organisms by enumerating the number of regions where several values above a specified value are continuous in the likelihood spectrum.

6. The sensor according to any one of claims 1 to 4, The second number information calculation unit comprises a determination unit based on machine learning, which has been previously learned through a likelihood spectrum as teacher data; The second number information calculation unit inputs the likelihood spectrum to the determination unit and outputs the determination result as the number of people.

7. The sensor according to claim 6, The second number information calculation unit uses a convolutional neural network.

8. The sensor according to any one of claims 1 to 4, The second number information calculation unit comprises: A peak search unit that extracts, as a first peak set, a peak that is the maximum value within a specified range among the peaks in the likelihood spectrum. A false peak determination unit that, for each peak included in the first peak set, calculates the y% value of the values included in a specified range around the peak, and extracts, as a second peak set, peaks for which the difference between the y% value and the peak is equal to or greater than a specified threshold. A peak sorting unit sorts the peaks included in the above-described second peak set in descending order; and A ratio test unit calculates the ratio or difference between the i-th peak and the (i + 1)-th peak of the above-described second peak set sorted in descending order, and outputs the i for which the ratio or difference is the largest as the second number of people information.

9. The sensor according to any one of claims 1 to 4, wherein the likelihood spectrum calculation unit is a MUSIC spectrum calculation unit that uses MUSIC, i.e., the multiple signal classification method.

Images