Signal analysis device, control circuit, storage medium, and signal analysis method

Abstract

A signal analysis device (30) comprises: an FFT (31) that converts a signal based on a chirp signal, which is acquired from a receiver group, into a frequency domain signal from a time domain signal, and stores the resulting signal as an observation value in an observation value storage unit (32); a localization determination unit (35) that determines a localized domain in which voxels having a voxel reflectance indicating that there is a possibility that an object is present in an observation domain are localized, and generates localization information; a sparse product-sum unit (37) that, by using local index values corresponding to the localized region indicated by the localization information and a local observation mode vector value corresponding to a right singular vector for each of the local index values, reads an observation value corresponding to the local index value from the observation value storage unit (32), and that, by performing a product-sum operation of the local observation mode vector value and an observation value vector represented by the observation value, calculates a local feature amount vector; and an inference unit (38) that, by using the local feature amount vector, generates object detection information indicating the detection state of the object.

Description

Signal analysis device, control circuit, storage medium, and signal analysis method

[0001] The present disclosure relates to a signal analysis device, a control circuit, a storage medium, and a signal analysis method used in object detection.

[0002] Conventionally, an array antenna has been used to estimate the three-dimensional shape of an object through 3D-RF (Three Dimensional-Radio Frequency) imaging, thereby performing object detection. However, in order to obtain a high-resolution three-dimensional shape using a wideband RF (Radio Frequency) signal, it is necessary to solve a huge inverse problem of estimating the reflectance of a large number of lattice points representing the three-dimensional reflectance distribution from a large number of array received signals, which results in an extremely heavy signal processing load. To address this problem, Patent Document 1 discloses a technology that uses an optical camera to measure the rough location of an object and performs processing based on the measurement results, assuming that the reflectance distribution is sparse, in order to reduce the signal processing load.

[0003] US Patent Application Publication No. 2016 / 0291148

[0004] However, the above-described conventional technology has the problem that it is not possible to reduce the processing load of the imaging process itself.

[0005] The present disclosure has been made in view of the above, and aims to provide a signal analysis device that can reduce the processing load of object detection.

[0006] In order to solve the above-described problems and achieve the object, the signal analysis device of the present disclosure includes: a converter that converts signals based on the chirp signals acquired from a group of receivers made up of a plurality of receiving antennas, the chirp signals being synchronized and distributed to each transmitting antenna, into an observation area from the transmitters of a transmitting device having a group of transmitters made up of a plurality of transmitting antennas, and the signal converted from a time-domain signal into a frequency-domain signal, and stores the converted signals as observation values ​​in a memory unit; the memory unit; a localization determination unit that determines localization areas in the observation area where voxels with voxel reflectances indicating the possibility of the presence of an object are localized, and generates localization information; a local product-sum unit that reads out from the memory, using local index values ​​corresponding to the localization areas indicated in the localization information and local observation mode vector values ​​corresponding to right singular vectors for each local index value, and performs a product-sum operation on the local observation mode vector values ​​and the observation value vector represented by the observation values, to calculate a local feature vector; and an inference unit that generates object detection information indicating an object detection status using the local feature vector.

[0007] The signal analysis device according to the present disclosure has an effect of reducing the processing load of object detection.

[0008] 5 is a diagram showing an example of the configuration of an object detection system according to embodiment 1. FIG. 5 shows an example of the configuration of a signal analysis device provided in a receiving device according to embodiment 1. FIG. 6 shows a flowchart illustrating the operation of the signal analysis device according to embodiment 1. FIG. 6 shows an example of an observation model system in the object detection system according to embodiment 1. FIG. 7 shows an example of an image in voxel space of a left singular vector corresponding to a certain local observation vector in the object detection system according to embodiment 1. FIG. 8 shows an example of an image in voxel space of a left singular vector corresponding to a local observation vector different from that in FIG. 5 in the object detection system according to embodiment 1. FIG. 9 shows an example of the arrangement of an observation target in the observation model system of the object detection system according to embodiment 1. FIG. 10 shows an example of an image in voxel space shown from observation values ​​by an observation target in the observation model system of the object detection system according to embodiment 1. FIG. 2 illustrates an example of an image in voxel space shown from observation values ​​by an observation target in an observation model system of an object detection system according to embodiment 1. FIG. 3 illustrates an example of an image in voxel space shown from observation values ​​by an observation target in an observation model system of an object detection system according to embodiment 1. FIG. 4 illustrates an example of a configuration of a processing circuit that realizes a signal analysis device according to embodiment 1 when realized by a processor and memory. FIG. 5 illustrates an example of a processing circuit that realizes a signal analysis device according to embodiment 1 when configured with dedicated hardware. FIG. 6 illustrates an example of a configuration of an object detection system according to embodiment 2. FIG. 7 illustrates an example of a configuration of a signal analysis device provided in a receiving device according to embodiment 2. Flowchart showing the operation of the signal analysis device according to embodiment 2. FIG. 8 illustrates an example of a configuration of an object detection system according to embodiment 3. FIG. 9 illustrates an example of a configuration of a signal analysis device provided in a receiving device according to embodiment 3. Flowchart showing the operation of the signal analysis device according to embodiment 3.

[0009] A signal analysis device, a control circuit, a storage medium, and a signal analysis method according to embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0010] 1 is a diagram showing an example of the configuration of an object detection system 40 according to embodiment 1. The object detection system 40 includes a transmitting device 10 and a receiving device 20, and is a system that detects an object 50 in an observation area. The transmitting device 10 includes a transmitter group 11 consisting of a plurality of transmitting antennas. The receiving device 20 includes a receiver group 21 consisting of a plurality of receiving antennas, and a signal analysis device 30.

[0011] In FIG. 1, the transmitter group 11 of the transmitting device 10 emits a modulated high-frequency chirp signal. That is, the transmitter group 11 of the transmitting device 10 transmits synchronized high-frequency chirp signals distributed to each transmitting antenna into the observation area. The high-frequency chirp signal is emitted as a w-th frequency component with a field intensity s(k, w) to the k-th voxel located at the k-th coordinate (x(k), y(k), z(k)). It is reflected by the k-th voxel with a reflectivity a(k). This state is represented in FIG. 1 as f(k, w) = a(k) * s(k, w). The receiver group 21 of the receiving device 20 receives the signal as a received signal component g(r, w) at each r-th element. That is, the receiver group 21 of the receiving device 20 receives a portion of the high-frequency chirp signal reflected or scattered in the observation area. In the receiving device 20 , the signal analyzing device 30 processes the signals received by the receiver group 21 to detect the object 50 .

[0012] The indexes r, w, j, and k are defined by natural numbers Nr, Nw, Ng, and Nv, respectively, as rε1,2,...,Nr, wε1,2,...,Nw, jε1,2,...,Ng, and kε1,2,...,Nv. The relationship Ng=Nr·Nw holds.

[0013] The transmitting device 10 performs processes such as DA (Digital to Analog) conversion, modulation, and upconversion on a baseband signal or an intermediate frequency signal to generate a high-frequency chirp signal, which is then transmitted from the transmitter group 11. The receiving device 20 performs processes such as downconversion, demodulation, and AD (Analog to Digital) conversion on the received signal received by the receiver group 21, and the resulting baseband signal or intermediate frequency signal is input to the signal analyzing device 30. In the object detection system 40, the operation up to the input of the baseband signal or intermediate frequency signal based on the high-frequency chirp signal to the signal analyzing device 30 is similar to that of a general object detection system, and therefore a detailed description thereof will be omitted. In the following description, the baseband signal or intermediate frequency signal based on the high-frequency chirp signal input to the signal analyzing device 30 may be simply referred to as a signal based on the high-frequency chirp signal.

[0014] Examples of high-frequency chirp signals used in the object detection system 40 include, but are not limited to, step chirp signals, linear chirp signals, parabolic chirp signals, and exponential chirp signals. High-frequency chirp signals other than those listed above may also be used in the object detection system 40. In the following description, high-frequency chirp signals may be simply referred to as chirp signals.

[0015] Here, if the scattering source vector of dimension Nv / Nw is f, the observation vector of dimension Ng obtained by reception is g, and the observation matrix representing the linear mapping relationship between the scattering source vector f and the observation vector g is H, then it can be considered that there is a relationship expressed by equation (1).

[0016] g = Hf ... (1)

[0017] The scattering source vector f can be expressed in equation (2) by a voxel scattering attenuation vector a whose dimension is Nv, which is 1 / Nw of the dimension of the scattering source vector f, and an illumination matrix S, which is a sparse matrix whose rank is equal to Nv.

[0018] f = Sa ... (2)

[0019] Combining equations (1) and (2), the relationship in equation (3) is obtained.

[0020] g = HSa ... (3)

[0021] Here, the image observation matrix Q is defined as in equation (4).

[0022] HS≡Q ... (4)

[0023] Using equation (4), equation (3) can be expressed as equation (5).

[0024] g = Qa ... (5)

[0025] If the image observation matrix Q is regular, the inverse matrix of the image observation matrix Q can be obtained, and equation (5) can be expressed as equation (6).

[0026] a = Q －１ g … (6)

[0027] In formula (6), Q －１ is the inverse matrix of the image observation matrix Q. This allows the voxel scattering attenuation rate vector a to be directly calculated. However, since the image observation matrix Q is generally not a regular matrix or a square matrix, the Moore-Penrose pseudo-inverse matrix Q ＋ is obtained by singular value analysis, and the voxel scattering attenuation rate vector a can be expressed as in equation (7).

[0028] a = Q ＋ g … (7)

[0029] Here, the high-resolution three-dimensional voxel scattering attenuation vector has a very large number of elements, for example, Nv is 10 ９ Therefore, the imaging calculation of equation (7) imposes an extremely large processing load.

[0030] On the other hand, in a practical observation system with a limited number of elements, the image observation matrix Q will be irregular or non-square, as described above. The number of elements in the observation vector g corresponding to the image observation matrix Q will be significantly smaller than the number of elements in the voxel scattering attenuation vector a. Because the observation vector g corresponding to the image observation matrix Q is the origin of the voxel scattering attenuation vector a, it has an amount of information equal to or greater than that of the voxel scattering attenuation vector a. From the above, by selecting right singular vectors corresponding to large singular values, it is possible to obtain corresponding dominant feature vectors without performing imaging operations using huge left singular vectors with the same number of elements as the number of voxels.

[0031] Fig. 2 is a diagram showing an example configuration of a signal analyzing device 30 included in the receiving device 20 according to embodiment 1. The signal analyzing device 30 includes an FFT (Fast Fourier Transform) 31, an observed value storage unit 32, a global storage unit 33, a sparse product-sum unit 34, a locality determining unit 35, a local storage unit 36, a sparse product-sum unit 37, and an inference unit 38. Fig. 3 is a flowchart showing the operation of the signal analyzing device 30 according to embodiment 1.

[0032] The FFT 31 acquires, from the receiver group 21, a signal based on the high-frequency chirp signal received by the receiver group 21. The FFT 31 converts the signal based on the high-frequency chirp signal acquired from the receiver group 21 from a time domain signal to a frequency domain signal. The FFT 31 stores the converted frequency domain signal as an observation value in the observation value storage unit 32 (step S11). The observation value storage unit 32 stores the frequency domain signal converted by the FFT 31 as an observation value. In the following description, the observation value storage unit 32 may be simply referred to as a storage unit.

[0033] The global storage unit 33 stores the specified global index values ​​and the observation mode vector values ​​corresponding to the right singular vectors for each global index value. The global index values ​​and observation mode vector values ​​are stored in the global storage unit 33 in advance by, for example, a user of the object detection system 40 before operating the object detection system 40. The signal analysis device 30 can also be configured without the global storage unit 33 by storing the global index values ​​and observation mode vector values ​​in the sparse product-sum unit 34. The global index values ​​and the observation mode vector values ​​corresponding to the right singular vectors for each global index value are characterized by being selected in combination so that the overall reflectivity within each region for which localization is to be determined can be determined as independently as possible between the regions. Additionally, observation values ​​from multiple receivers that are close in time may be used to improve the accuracy of local determination, or information from other sensors, such as an optical camera or an ultrasonic sensor, may be used to improve the accuracy of local determination.

[0034] The sparse product-sum unit 34 reads out the global index values ​​and the observation mode vector values ​​corresponding to the right singular vectors for each global index value from the global storage unit 33. The sparse product-sum unit 34 uses the global index values ​​to read out the observation values ​​corresponding to the global index values ​​from the observation value storage unit 32. The sparse product-sum unit 34 uses the observation mode vector values ​​to perform a product-sum operation on the observation mode vector values ​​and the observation value vectors represented by the observation values ​​to calculate vector inner product values ​​(step S12). The sparse product-sum unit 34 outputs the vector inner product values ​​to the locality determination unit 35. In the following description, the sparse product-sum unit 34 may be referred to as a global product-sum unit.

[0035] The localization determination unit 35 determines a localized region where voxels with voxel reflectance indicating the possibility of the presence of an object 50 are localized in the observation region, and generates localization information (step S13). In the first embodiment, the localization determination unit 35 generates the localization information using the vector dot product value obtained from the sparse product-sum unit 34. Specifically, the localization determination unit 35 determines a coarse region where voxel reflectance is localized based on the vector dot product value obtained from the sparse product-sum unit 34, and generates localization information indicating the determination result. The localization determination unit 35 stores the localization information in the local storage unit 36. The local storage unit 36 ​​is a storage unit that stores the localization information generated by the localization determination unit 35. The local storage unit 36 ​​also stores local index values ​​and local observation mode vector values ​​corresponding to the right singular vectors for each local index value. The local index values ​​and the local observation mode vector values ​​corresponding to the right singular vectors for each local index value are selected to include observation modes that are highly sensitive to feature quantities corresponding to high spatial frequency components in the reflectance distribution in the local region of interest. In addition, for local regions where the reflectance distribution is estimated to be low based on the detected global index value, it can be estimated that the degree of erroneous detection due to received signal components from that region is low, and therefore observation performance can be improved by using more observation modes rather than limiting them to those with low sensitivity to that region.

[0036] Based on the locality information stored in the local storage unit 36, the sparse product-sum unit 37 reads, from the local storage unit 36, pairs of local index values ​​corresponding to regions corresponding to the locality information and local observation mode vector values ​​corresponding to the right singular vectors for each local index value. The sparse product-sum unit 37 reads observation values ​​corresponding to the local index values ​​from the observation value storage unit 32. The sparse product-sum unit 37 calculates vector dot products by performing a product-sum operation on the local observation mode vector values ​​and the observation value vectors represented by the observation values. The sparse product-sum unit 37 outputs the calculated vector dot product value to the inference unit 38 as a local feature vector. In this way, the sparse product-sum unit 37 reads, from the local storage unit 36, observation values ​​corresponding to the local index values ​​using the local index values ​​corresponding to the local region indicated by the locality information and the local observation mode vector values ​​corresponding to the right singular vectors for each local index value, and calculates local feature vectors by performing a product-sum operation on the local observation mode vector values ​​and the observation value vectors represented by the observation values ​​(step S14). In the following description, the sparse product-sum unit 37 may be referred to as a local product-sum unit. Note that the signal analyzing device 30 may be configured without the local storage unit 36 ​​by storing the local index values ​​and the local observation mode vector values ​​in the sparse product-sum unit 37 and having the locality determining unit 35 output the locality information directly to the sparse product-sum unit 37.

[0037] The inference unit 38 uses the local feature vectors acquired from the sparse product-sum unit 37 to generate and output object detection information indicating the detection status of the object 50 (step S15). Specifically, the inference unit 38 processes the local feature vectors acquired from the sparse product-sum unit 37 by deep learning using a CNN (Convolutional Neural Network), i.e., a convolutional neural network, to generate the object detection information. Note that the method by which the inference unit 38 generates the object detection information is not limited to this. The inference unit 38 may also process the local feature vectors acquired from the sparse product-sum unit 37 by machine learning to generate the object detection information.

[0038] FIG. 4 is a diagram showing an example of an observation model system in an object detection system 40 according to the first embodiment. In FIG. 4, the fine point clouds arranged in a cubic shape are voxels. Also, in FIG. 4, "+" indicates each receiving element of the receiver group 21 provided in the receiving device 20. FIG. 4 shows that an 8x8 receiving element group, i.e., the receiver group 21, is arranged on two surfaces around the voxels.

[0039] Fig. 5 is a diagram illustrating an example of a left singular vector corresponding to a certain local observation vector in voxel space in object detection system 40 according to embodiment 1. Fig. 6 is a diagram illustrating an example of a left singular vector corresponding to a different local observation vector from that in Fig. 5 in object detection system 40 according to embodiment 1, as an image in voxel space. It can be seen from Figs. 5 and 6 that localization information in voxel space can be extracted using the local observation vector.

[0040] FIG. 7 is a diagram illustrating an example of the arrangement of an observation target in the observation model system of object detection system 40 according to embodiment 1. FIG. 8 is a first diagram illustrating an example of an image in voxel space represented by observation values ​​of an observation target in the observation model system of object detection system 40 according to embodiment 1. FIG. 9 is a second diagram illustrating an example of an image in voxel space represented by observation values ​​of an observation target in the observation model system of object detection system 40 according to embodiment 1. FIG. 10 is a third diagram illustrating an example of an image in voxel space represented by observation values ​​of an observation target in the observation model system of object detection system 40 according to embodiment 1. FIG. 8 illustrates an example of an image in voxel space represented by observation values ​​of an observation target in the observation model system of object detection system 40 according to embodiment 1, as a bird's-eye perspective view from the z-axis direction, using all observation mode vectors of singular value decomposition. FIG. 9 illustrates an example of an image in voxel space represented by observation mode vectors that are 1 / 17 of the observation mode vectors of singular value decomposition, as a bird's-eye perspective view from the z-axis direction. Figure 10 shows an example of an image in voxel space using 1 / 34 observation mode vectors of singular value decomposition as a bird's-eye view from the z-axis direction. Figures 8, 9, and 10 show that the amount of information is well preserved even when the amount of calculation is reduced using sparse product-sum.

[0041] Next, the hardware configuration of the signal analyzing device 30 will be described. In the signal analyzing device 30, the observed value storage unit 32, the global storage unit 33, and the local storage unit 36 ​​are memories. The sparse product-sum unit 34, the locality determination unit 35, the sparse product-sum unit 37, and the inference unit 38 are realized by processing circuits. The processing circuit may be a processor and memory that executes a program stored in a memory, or may be dedicated hardware. The processing circuit is also called a control circuit.

[0042] FIG. 11 is a diagram illustrating an example of the configuration of a processing circuit 90 that implements the signal analyzing device 30 according to the first embodiment when the processing circuit is implemented by a processor 91 and a memory 92. The processing circuit 90 illustrated in FIG. 11 is a control circuit and includes a processor 91 and a memory 92. When the processing circuit 90 is configured with the processor 91 and the memory 92, each function of the processing circuit 90 is implemented by software, firmware, or a combination of software and firmware. The software or firmware is written as a program and stored in the memory 92. The processing circuit 90 implements each function by having the processor 91 read and execute the program stored in the memory 92. That is, the processing circuit 90 includes the memory 92 for storing a program that results in the processing of the signal analyzing device 30 being executed. This program can also be said to be a program that causes the signal analyzing device 30 to execute each function implemented by the processing circuit 90. This program may be provided by a storage medium on which the program is stored, or by other means such as a communication medium.

[0043] The program includes a conversion step in which an FFT 31 converts a signal based on a high frequency chirp signal acquired from a group of receivers 21 consisting of a plurality of receiving antennas, which receives a portion of the high frequency chirp signal reflected or scattered in the observation area from a group of transmitters 11 of a transmitting device 10 that has a group of transmitters 11 consisting of a plurality of transmitting antennas, from a time domain signal to a frequency domain signal, and stores the signal as an observation value in an observation value storage unit 32; and a localization determination unit 35 determines a localized region in which voxels having voxel reflectances indicating the possibility of the presence of an object 50 in the observation area are localized, and stores the localization information. a local product-sum step in which the sparse product-sum unit 37 reads out observation values ​​corresponding to the local index values ​​from the observation value storage unit 32 using local index values ​​corresponding to the local region indicated by the localization information and local observation mode vector values ​​corresponding to the right singular vectors for each local index value, and calculates local feature vectors by performing a product-sum operation on the local observation mode vector values ​​and the observation value vectors represented by the observation values; and an inference step in which the inference unit 38 generates object detection information indicating the detection status of the object 50, using the local feature vectors.

[0044] Here, the processor 91 is, for example, a CPU (Central Processing Unit), a processing device, an arithmetic unit, a microprocessor, a microcomputer, or a DSP (Digital Signal Processor), etc. The memory 92 is, for example, a non-volatile or volatile semiconductor memory such as a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable ROM), or an EEPROM (Electrically EPROM), a magnetic disk, a flexible disk, an optical disk, a compact disk, a minidisk, or a DVD (Digital Versatile Disc).

[0045] FIG. 12 is a diagram illustrating an example of a processing circuit 93 that implements the signal analyzing device 30 according to the first embodiment and is configured with dedicated hardware. The processing circuit 93 illustrated in FIG. 12 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination thereof. The processing circuit may be partially implemented with dedicated hardware and partially implemented with software or firmware. In this way, the processing circuit can implement each of the above-described functions using dedicated hardware, software, firmware, or a combination thereof.

[0046] As described above, according to this embodiment, in the signal analysis device 30, the FFT 31 converts signals based on high-frequency chirp signals acquired from the receiver group 21 from time-domain signals to frequency-domain signals and stores the converted signals as observation values ​​in the observation value storage unit 32. The sparse product-sum unit 34 reads observation values ​​corresponding to global index values ​​from the observation value storage unit 32 and performs a product-sum operation on the observation mode vector values ​​and the observation value vectors to calculate vector inner products. The locality determination unit 35 generates locality information using the vector inner product values. The sparse product-sum unit 37 reads observation values ​​corresponding to local index values ​​from the local storage unit 36 ​​and performs a product-sum operation on the local observation mode vector values ​​and the observation value vectors to calculate local feature vectors. The inference unit 38 generates and outputs object detection information using the local feature vectors.

[0047] As a result, the signal analysis device 30 can obtain high-order features required for object detection using a convolutional neural network or machine learning, i.e., local feature vectors that are localized features, without performing high-resolution 3D-RF imaging, thereby enabling efficient object detection and reducing the processing load of object detection. Furthermore, by reducing the processing load, the signal analysis device 30 can reduce the memory resources required to store mode data. Furthermore, for the same level of object detection estimation accuracy, the signal analysis device 30 can achieve the effect of speeding up signal processing by reducing the processing load. Furthermore, by reducing the processing load, the signal analysis device 30 can achieve the effect of improving object detection estimation accuracy compared to signal analysis devices with the same level of memory resources.

[0048] Second Embodiment In the first embodiment, the locality determining unit 35 generates locality information using the vector inner product value calculated by the sparse product-sum unit 34. In the second embodiment, a case will be described in which the locality determining unit generates locality information using different information.

[0049] Fig. 13 is a diagram showing an example configuration of an object detection system 40a according to embodiment 2. The object detection system 40a is obtained by replacing the receiving device 20 with a receiving device 20a in the object detection system 40 according to embodiment 1 shown in Fig. 1. The receiving device 20a includes a receiver group 21, a camera 22, and a signal analysis device 30a. The camera 22 captures an image of an observation area. The camera 22 outputs image information obtained by capturing the image of the observation area to the signal analysis device 30a.

[0050] Fig. 14 is a diagram showing an example of the configuration of a signal analyzing device 30a included in a receiving device 20a according to embodiment 2. The signal analyzing device 30a is obtained by deleting the global storage unit 33, the sparse product-sum unit 34, and the locality determining unit 35 from the signal analyzing device 30 according to embodiment 1 shown in Fig. 2 and adding a locality determining unit 35a. Fig. 15 is a flowchart showing the operation of the signal analyzing device 30a according to embodiment 2.

[0051] In the signal analyzing device 30a, the operation of step S11 is the same as the operation of step S11 in the flowchart of the first embodiment shown in Figure 3. The localization determining unit 35a acquires image information from the camera 22 that captures an observation area (step S21). The localization determining unit 35a generates localization information using the image information acquired from the camera 22 (step S22). In the signal analyzing device 30a, the operations of subsequent steps S14 and S15 are the same as the operations of steps S14 and S15 in the flowchart of the first embodiment shown in Figure 3.

[0052] As described above, according to this embodiment, in the signal analyzing device 30a, the localization determining unit 35a generates localization information using image information acquired from the camera 22 that captures the observation area. Even in this case, the signal analyzing device 30a can obtain the same effects as the signal analyzing device 30 of the first embodiment.

[0053] Embodiment 3 In the first embodiment, the locality determination unit 35 generates locality information using the vector dot product values ​​calculated by the sparse product-sum unit 34. In the second embodiment, the locality determination unit 35a generates locality information using image information acquired from the camera 22. In the third embodiment, a case will be described in which the locality determination unit generates locality information using the vector dot product values ​​calculated by the sparse product-sum unit 34 and the image information acquired from the camera 22.

[0054] Fig. 16 is a diagram showing an example configuration of an object detection system 40b according to embodiment 3. The object detection system 40b is obtained by replacing the receiving device 20 with a receiving device 20b in the object detection system 40 according to embodiment 1 shown in Fig. 1. The receiving device 20b includes a receiver group 21, a camera 22, and a signal analysis device 30b. The camera 22 captures an image of an observation area. The camera 22 outputs image information obtained by capturing an image of the observation area to the signal analysis device 30b.

[0055] Fig. 17 is a diagram showing an example of the configuration of a signal analyzing device 30b included in a receiving device 20b according to embodiment 3. The signal analyzing device 30b is obtained by replacing the localization determining unit 35 with a localization determining unit 35b in the signal analyzing device 30 according to embodiment 1 shown in Fig. 2. Fig. 18 is a flowchart showing the operation of the signal analyzing device 30b according to embodiment 3.

[0056] In the signal analyzing device 30b, the operations of steps S11 and S12 are the same as those of steps S11 and S12 in the flowchart of the first embodiment shown in Fig. 3. Furthermore, the operation of step S21 is the same as that of step S21 in the flowchart of the second embodiment shown in Fig. 15. The localization determining unit 35b generates localization information using the vector dot product value acquired from the sparse product-sum unit 34 and image information acquired from the camera 22 (step S31). In the signal analyzing device 30b, the operations of subsequent steps S14 and S15 are the same as those of steps S14 and S15 in the flowchart of the first embodiment shown in Fig. 3.

[0057] As described above, according to this embodiment, in the signal analyzing device 30b, the localization determining unit 35b generates localization information using the vector inner product value calculated by the sparse product-sum unit 34 and image information acquired from the camera 22 that captures the observation area. Even in this case, the signal analyzing device 30b can obtain the same effects as the signal analyzing device 30 of the first embodiment.

[0058] The configurations shown in the above embodiments are merely examples, and may be combined with other known technologies, or different embodiments may be combined with each other. It is also possible to omit or modify parts of the configurations as long as they do not deviate from the gist of the invention.

[0059] 10 Transmitting device, 11 Transmitter group, 20, 20a, 20b Receiving device, 21 Receiver group, 22 Camera, 30, 30a, 30b Signal analysis device, 31 FFT, 32 Observation value memory unit, 33 Global memory unit, 34, 37 Sparse product sum unit, 35, 35a, 35b Locality determination unit, 36 Local memory unit, 38 Inference unit, 40, 40a, 40b Object detection system, 50 Object, 90, 93 Processing circuit, 91 Processor, 92 Memory.

Claims

1. A transmitter comprising a group of transmitters consisting of multiple transmitting antennas transmits a synchronized chirp signal distributed to each transmitting antenna from the group of transmitters to an observation area, and a converter converts a signal based on the chirp signal obtained from a group of receivers consisting of multiple receiving antennas that receive a portion of the chirp signal reflected or scattered in the observation area into a frequency domain signal and stores it in a memory unit as an observed value. The aforementioned storage unit, A localization determination unit determines the localized region where voxes of the voxel reflectance, which indicate the possibility of an object being present in the aforementioned observation region, are localized, and generates localization information. A local sum-of-products unit calculates a local feature vector by reading the observed values ​​corresponding to the localized region indicated by the locality information and the local observation mode vector values ​​corresponding to the right singular vector for each local index value from the storage unit, and performing a sum-of-products operation between the local observation mode vector values ​​and the observed value vector represented by the observed values. An inference unit that generates object detection information indicating the detection status of the object using the local feature vector, A signal analysis device characterized by being equipped with the following features.

2. A global sum-of-products unit reads the observed values ​​corresponding to the global index values ​​from the storage unit using a defined global index value and an observed value vector corresponding to the right singular vector for each global index value, and calculates the intravector product by performing a sum-of-products operation between the observed mode vector value and the observed value vector represented by the observed value. Equipped with, The localization determination unit generates the localization information using the vector intraproduct value. The signal analysis device according to feature 1.

3. The localization determination unit generates the localization information using image information acquired from a camera that photographs the observation area. The signal analysis device according to feature 1.

4. A global sum-of-products unit reads the observed values ​​corresponding to the global index values ​​from the storage unit using a defined global index value and an observed value vector corresponding to the right singular vector for each global index value, and calculates the intravector product by performing a sum-of-products operation between the observed mode vector value and the observed value vector represented by the observed value. Equipped with, The localization determination unit generates the localization information using the vector inner product and image information acquired from a camera capturing the observation region. The signal analysis device according to feature 1.

5. The inference unit processes the local feature vectors using deep learning with a convolutional neural network to generate the object detection information. The signal analysis device according to feature 1.

6. The inference unit processes the local feature vector using machine learning to generate the object detection information. The signal analysis device according to feature 1.

7. The chirp signal is a step chirp signal, a linear chirp signal, a parabolic chirp signal, or an exponential chirp signal. A signal analysis device according to any one of claims 1 to 6.

8. A control circuit for controlling a signal analysis device, A transmitting device comprising a group of transmitters consisting of multiple transmitting antennas transmits a synchronized chirp signal distributed to each transmitting antenna from the group of transmitters to the observation area. A signal based on the chirp signal, obtained from a group of receivers consisting of multiple receiving antennas that receive a portion of the chirp signal reflected or scattered in the observation area, is converted from a time-domain signal to a frequency-domain signal and stored as an observed value in a storage unit. The localized region where voxes in the voxel reflectance, which indicate the possibility of an object being present in the aforementioned observation area, are localized is determined, and localization information is generated. Using the local index value corresponding to the localized region indicated by the localization information and the local observation mode vector value corresponding to the right singular vector for each local index value, the observed value corresponding to the local index value is read from the storage unit, and a sum-of-products operation is performed between the local observation mode vector value and the observed value vector represented by the observed value to calculate the local feature vector. Using the aforementioned local feature vector, object detection information indicating the detection status of the object is generated. A control circuit characterized by causing the signal analysis device to perform the above-mentioned operation.

9. A storage medium in which a program for controlling a signal analysis device is stored, The aforementioned program, A transmitting device comprising a group of transmitters consisting of multiple transmitting antennas transmits a synchronized chirp signal distributed to each transmitting antenna from the group of transmitters to the observation area. A signal based on the chirp signal, obtained from a group of receivers consisting of multiple receiving antennas that receive a portion of the chirp signal reflected or scattered in the observation area, is converted from a time-domain signal to a frequency-domain signal and stored as an observed value in a storage unit. The localized region where voxes in the voxel reflectance, which indicate the possibility of an object being present in the aforementioned observation area, are localized is determined, and localization information is generated. Using the local index value corresponding to the localized region indicated by the localization information and the local observation mode vector value corresponding to the right singular vector for each local index value, the observed value corresponding to the local index value is read from the storage unit, and a sum-of-products operation is performed between the local observation mode vector value and the observed value vector represented by the observed value to calculate the local feature vector. Using the aforementioned local feature vector, object detection information indicating the detection status of the object is generated. A storage medium characterized by having the signal analysis device perform the aforementioned action.

10. A converter performs a conversion step in which a chirp signal, which is synchronized and distributed to each transmitting antenna from a group of transmitters of a transmitting device comprising a group of transmitting antennas, is transmitted to an observation area, and a signal based on the chirp signal, which is received by a group of receiving antennas comprising a group of receiving antennas that receive a portion of the chirp signal reflected or scattered in the observation area, is converted from a time-domain signal to a frequency-domain signal and stored in a storage unit as an observed value. Localization determination step: The localization determination unit determines the localized region where voxes of the voxel reflectance, which indicate the possibility of an object being present in the observation region, are localized, and generates localization information. The local sum-of-products unit performs a local sum-of-products step in which it reads the observed values ​​corresponding to the local index values ​​from the storage unit using the local index values ​​corresponding to the localized region indicated by the locality information and the local observation mode vector values ​​corresponding to the right singular vector for each local index value, and calculates a local feature vector by performing a sum-of-products operation between the local observation mode vector values ​​and the observed value vector represented by the observed values. The inference unit performs an inference step in which it generates object detection information indicating the detection status of the object using the local feature vector, A signal analysis method characterized by including the following.

11. The global sum-of-products unit reads the observed values ​​corresponding to the global index values ​​from the storage unit using the defined global index values ​​and the observed mode vector values ​​corresponding to the right singular vectors for each global index value, and performs a sum-of-products operation between the observed mode vector values ​​and the observed value vector represented by the observed values ​​to calculate the intravector product in the global sum-of-products step. Includes, In the localization determination step, the localization determination unit generates the localization information using the vector inner product. The signal analysis method according to feature 10.

12. In the localization determination step, the localization determination unit generates the localization information using image information acquired from a camera that photographs the observation area. The signal analysis method according to feature 10.

13. The global sum-of-products unit reads the observed values ​​corresponding to the global index values ​​from the storage unit using the defined global index values ​​and the observed mode vector values ​​corresponding to the right singular vectors for each global index value, and performs a sum-of-products operation between the observed mode vector values ​​and the observed value vector represented by the observed values ​​to calculate the intravector product in the global sum-of-products step. Includes, In the localization determination step, the localization determination unit generates the localization information using the vector inner product and image information acquired from a camera capturing the observation area. The signal analysis method according to feature 10.

14. In the inference step, the inference unit processes the local feature vector using deep learning with a convolutional neural network to generate the object detection information. The signal analysis method according to feature 10.

15. In the inference step, the inference unit processes the local feature vector by machine learning to generate the object detection information. The signal analysis method according to feature 10.

16. The chirp signal is a step chirp signal, a linear chirp signal, a parabolic chirp signal, or an exponential chirp signal. The signal analysis method according to any one of claims 10 to 15.