JP6673030B2 - Information processing apparatus, information processing method, and program - Google Patents

Description

  The present invention relates to an information processing device, an information processing method, and a program.

  2. Description of the Related Art In recent years, in ITS (Intelligent Transport Systems), road monitoring systems, and the like, various methods for identifying the type of traveling vehicle have been proposed.

  For example, Patent Literature 1 discloses a technique for identifying a vehicle type by comparing the intensity and duration of a reflected wave received when a traveling vehicle passes through a detection area with a model pattern stored in advance. ing.

  In addition, for example, in Patent Document 2, the length and height of the vehicle obtained from the duration of the intensity change of the reflected wave received when the traveling vehicle passes through the detection area are stored in advance for each vehicle type data. A technology for discriminating a vehicle type by comparing with a vehicle type is disclosed.

  Further, for example, Patent Literature 3 discloses a technique of calculating the height of a target from the ground using a vehicle-mounted radar having an array antenna.

JP-A-11-272988 JP 2014-99123 A JP 2014-52187 A

  However, when the technology described in Patent Document 1 is used, it is difficult to set a wide detection area in order to measure the duration of the reflected wave. Furthermore, as a radar installation condition, it is required that the transmitting / receiving antenna be directed to the side of the traveling lane of the vehicle. Therefore, for example, when two vehicles traveling in different lanes pass through the detection region at the same time, the intensity and duration of the reflected waves are combined for two vehicles, and there is a possibility that the vehicle type cannot be correctly identified. high.

  In addition, when the technology described in Patent Document 2 is used, a column for fixing the radar is required in order to install the radar vertically on the vehicle, so that the installation cost tends to be high. is there.

  In addition, when the technique described in Patent Document 3 is used, it is necessary to use a radar having an array antenna. However, such a radar generally has a complicated circuit. For this reason, strict noise countermeasures are required, and there is a problem that component devices are expensive.

  Therefore, the present invention has been made in view of the above-described problems, and an object of the present invention is to provide a new and improved information processing method capable of realizing high-precision object marking at a lower cost. An object is to provide an apparatus, an information processing method, and a program.

  In order to solve the above problems, according to an aspect of the present invention, a first hierarchy that abstracts a two-dimensional pattern shape of a feature amount extracted based on a frequency spectrum of a detected target, A multi-layer neural network including a second layer that determines an identification class of the target based on analysis information of the target candidate detected from the frequency spectrum of the target and an output from the first layer A target marker discriminating unit that identifies the target by using at least one of a partial spectrogram related to an amplitude value or a partial spectrogram related to a distance value extracted from a spectrogram based on the frequency spectrum. An information processing device including one of them is provided.

  The analysis information of the target candidate may include at least one of amplitude intensity, speed information, distance information, and angle information.

  The information processing apparatus may further include a feature extraction unit configured to extract data near the target candidate as the feature amount from a spectrogram obtained by expanding the frequency spectrum in a time direction.

  The information processing apparatus may further include a target detection unit that detects at least one or more target candidates based on the intensity of the frequency spectrum.

  The information processing apparatus further includes a target tracking unit that outputs a tracking result indicating a temporal correspondence relation between the plurality of target candidates, and the target marker unit is output by the multilayer neural network. The target may be identified based on the identification class and the tracking result.

  The target marking unit may identify the target based on the most frequently occurring identification class among the plurality of identification classes corresponding to the tracking result.

  When the frequency difference between two or more different target candidates exceeds a predetermined threshold, the target marker separate unit may not use the identification class related to the target candidate for the identification of the target.

  According to another embodiment of the present invention, there is provided a computer configured to abstract a two-dimensional pattern shape of a feature amount extracted based on a frequency spectrum of a detected target. A first hierarchy, and a second hierarchy that determines an identification class of the target based on analysis information of the target candidate detected from the frequency spectrum of the target and an output from the first hierarchy. A target marker discriminating unit that identifies the target by using a multilayer neural network including: the feature amount is a partial spectrogram related to an amplitude value or a partial spectrogram related to a distance value extracted from a spectrogram based on the frequency spectrum. And a program for functioning as an information processing device including at least one of them.

  According to another embodiment of the present invention, there is provided an information processing method using a multilayer neural network, wherein a feature amount extracted based on a frequency spectrum of a detected target is provided. Abstracting the two-dimensional pattern shape of the target, and analyzing the target candidate based on the analysis information of the target candidate detected from the frequency spectrum of the target and the abstracted two-dimensional pattern of the feature amount. Determining the discrimination class of the information processing, wherein the feature amount includes at least one of a partial spectrogram related to an amplitude value or a partial spectrogram related to a distance value extracted from a spectrogram based on the frequency spectrum. A method is provided.

  As described above, according to the present invention, it is possible to realize high-precision object labeling at lower cost.

It is an explanatory view for explaining an outline of an embodiment of the present invention. FIG. 3 is a functional block diagram of the information processing apparatus according to the embodiment. FIG. 9 is an explanatory diagram for describing detection of a target candidate by the target detection unit according to the embodiment; FIG. 3 is a diagram illustrating an example of a data structure of a feature amount extracted by a feature extraction unit according to the embodiment. FIG. 3 is a diagram illustrating an example of a network structure of the multilayer neural network according to the embodiment. It is an example of the tracking result output by the target tracking unit according to the embodiment. It is a flowchart showing a flow of processing of the information processing apparatus according to the same embodiment. 2 is a hardware configuration example of an information processing apparatus according to the present invention.

  Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the specification and the drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant description is omitted.

<1. Background>
2. Description of the Related Art In recent years, a road monitoring system such as ITS uses a technology for identifying a type of a traveling vehicle using various sensors. Examples of the above technology include vehicle type identification by image recognition processing. However, the image sensor is generally susceptible to differences in sunshine during the day or at night, and to weather such as rain and snow. For this reason, when used for continuous monitoring outdoors, means for eliminating the above-mentioned effects are required.

  On the other hand, as described in Patent Literature 1 and Patent Literature 2 described above, a vehicle type identification method using a microwave radar is also known. Microwave radar has a property that it is hard to be affected by sunshine and weather, and thus has an advantage over an image sensor in applications requiring environmental resistance.

  However, in the vehicle type identification using the microwave radar, it is required to install the microwave radar at a high place such as a bridge, and the installation cost tends to increase.

  An information processing apparatus, an information processing method, and a program according to an embodiment of the present invention have been conceived by paying attention to the above points. By installing a microwave radar on the roadside, the installation cost is significantly reduced. It is possible to reduce. In addition, the information processing apparatus, the information processing method, and the program according to an embodiment of the present invention can realize highly accurate vehicle type identification by employing a multilayer neural network. In the embodiments of the present invention described below, the effects of the features will be described while citing the structural features of the information processing apparatus according to the present invention.

<2. Embodiment>
[2.1. Overview of the present embodiment]
First, an outline of an embodiment of the present invention will be described with reference to FIG. FIG. 1 is an explanatory diagram for explaining the outline of the present embodiment. Referring to FIG. 1, an information processing apparatus 10 according to the present embodiment acquires a reaction from a target O1 such as a running vehicle by a microwave radar. At this time, the information processing device 10 according to the present embodiment may be set on the roadside as shown in FIG.

  The information processing apparatus 10 according to the present embodiment may use, for example, a CW (Continuous Wave; continuous wave) type microwave radar having an antenna with narrow directivity.

  As shown in FIG. 1, the information processing apparatus 10 according to the present embodiment can acquire, as a reception signal, a reflected wave generated when a microwave transmitted from a microwave radar comes into contact with a target O1. At this time, due to the Doppler effect, various effects related to the target O1 can be observed in the received signal. Specifically, the signal intensity of the received signal changes depending on the surface area and the distance of the target O1, and the frequency changes depending on the speed of the target O1.

  Further, when the information processing apparatus 10 uses a plurality of transmission waves having different oscillation frequencies, it is possible to measure the distance between the antenna and the target O1. Furthermore, if the positional relationship between the lane and the radar is clear in advance, the angle between the antenna and the target O1 can be estimated from the intensity and the distance related to the target O1. That is, the information processing apparatus 10 according to the present embodiment can acquire information on the intensity, speed, distance, and angle of the target O1 using the microwave radar.

  The information processing apparatus 10 according to the present embodiment realizes the identification of the target O1 with high accuracy by inputting the information obtained as described above and various types of information that can be analyzed from the information to the multilayer neural network. It is possible to In the following description, the functional configuration and processing of the information processing device 10 according to the present embodiment will be described in detail.

[2.2. Functional configuration example of information processing apparatus 10]
Next, an example of a functional configuration of the information processing apparatus 10 according to the present embodiment will be described with reference to FIG. FIG. 2 is a functional block diagram of the information processing apparatus 10 according to the present embodiment. Referring to FIG. 2, the information processing apparatus 10 according to the present embodiment includes a radar 110, a signal processing unit 120, a target detection unit 130, a feature extraction unit 140, a target tracking unit 150, and a target marking unit 160. . Hereinafter, each of the above configurations will be described in detail.

(Radar 110)
The radar 110 has a function of transmitting a radio wave such as a millimeter wave toward a target and receiving a reflected wave. Therefore, the radar 110 according to the present embodiment may be, for example, a CW-type radar having a single receiving element. Further, the radar 110 according to the present embodiment has a function of switching a plurality of oscillation frequencies in a stepwise or continuous manner and transmitting.

(Signal processing unit 120)
The signal processing unit 120 has a function of analyzing the frequency of the signal received by the radar 110 and calculating a frequency spectrum related to the Doppler frequency. The signal processing unit 120 according to the present embodiment can calculate the speed v [m / s] and the distance r [m] of the target.

  The signal processing unit 120 may calculate the speed v [m / s] of the target using, for example, the following equation (1).

Here, λ [m] in the above equation (1) indicates the wavelength of the transmission wave. Further, the Doppler frequency f _d [Hz] is calculated by using a fast Fourier transform (FFT) for a received signal received by the radar 110.

  In addition, the signal processing unit 120 may calculate the distance r [m] related to the target using, for example, the following equation (2). At this time, as described above, the radar 110 may have a function of switching a plurality of oscillation frequencies stepwise or continuously and transmitting.

Here, Δf _s [Hz] in the above equation (2) indicates a difference between transmission frequencies, and Δφ _s [rad] indicates a signal phase difference between the transmission frequencies. Further, c [m / s] in the above equation (2) is the speed of light.

  As described above, the signal processing unit 120 according to the present embodiment can acquire the frequency spectrum related to the Doppler frequency using the fast Fourier transform.

(Target detection unit 130)
The target detection unit 130 has a function of detecting at least one or more target candidates based on the intensity of the frequency spectrum acquired by the signal processing unit 120. Here, the target candidate according to the present embodiment may indicate a specific frequency that is a peak detection result of a frequency spectrum. Details of target candidate detection by the target detection unit 130 according to the present embodiment will be described later.

(Feature extraction unit 140)
The feature extraction unit 140 has a function of extracting a feature amount related to a target candidate detected by the target detection unit 130. Here, the feature quantity according to the present embodiment may include at least one of a partial spectrogram related to an amplitude value and a partial spectrogram related to a distance value extracted from a spectrogram based on a frequency spectrum. The feature extraction unit 140 according to the present embodiment extracts, as a feature amount, data near the target candidate from a spectrogram obtained by expanding the frequency spectrum of the target candidate in the time direction. Details of the feature amount extraction by the feature extraction unit 140 according to the present embodiment will be described later.

(Target tracking unit 150)
The target tracking unit 150 has a function of outputting a tracking result indicating a temporal correspondence relationship between a plurality of target candidates detected by the target detection unit 130. The target tracking unit 150 according to the present embodiment can output the above tracking result by considering the consistency of the speed and the distance of the target candidate.

  Specifically, the target tracking unit 150 according to the present embodiment tracks target candidates output from the target detection unit 130 at arbitrary time intervals. The target tracking unit 150 according to the present embodiment, for example, when associating the target candidate at a certain time k with the target candidate at the next time k + 1, at the speed related to the target candidate at the time k, at the time k + 1 The processing may be performed such that an error from the speed of the plurality of detected (i = 1,..., N) target candidates is minimized. Specifically, it is expressed by the following equation (3).

  Note that the above equation (3) shows a case where the s-th target candidate at time k + 1 is associated with the target candidate at time k. The target tracking unit 150 according to the present embodiment can track the target candidate by repeating the above operation for each time. Note that the tracking result output by the target tracking unit 150 according to the present embodiment may be used for identification by the target marking unit 160 described later. The details of each object marker using the tracking result according to the present embodiment will be described later.

(Object marker separate section 160)
The object marker classification unit 160 has a function of performing object marker classification based on a machine learning model using the feature amount extracted by the feature extraction unit 140. For this reason, the object marker distinction unit 160 according to the present embodiment includes a first layer that abstracts a two-dimensional pattern shape of a feature amount extracted based on the frequency spectrum of the detected object, A multi-layer neural network including a second layer that determines the identification class of the target based on the analysis information of the target candidate detected from the frequency spectrum of the target and the output from the first layer. Is fine.

  Here, the analysis information of the target candidate may include at least one of amplitude intensity, speed information, distance information, and angle information.

  In addition, the target marking unit 160 according to the present embodiment has a function of identifying a target based on the identification class output by the multilayer neural network and the tracking result output by the target tracking unit 150. You may.

  At this time, the object sign classification unit 160 according to the present embodiment can identify the target based on the most frequently occurring identification class among the plurality of identification classes corresponding to the tracking result.

  When the frequency difference between two or more different target candidates exceeds a predetermined threshold, the target marker classification unit 160 according to the present embodiment uses the identification class for the target candidate for target identification. Is one of the features. The details of each object label by the object marker separate unit 160 according to the present embodiment will be described later.

[2.3. Details of target candidate detection]
Next, details of target candidate detection according to the present embodiment will be described. As described above, the target detection unit 130 according to the present embodiment has a function of detecting at least one or more target candidates based on the intensity of the frequency spectrum acquired by the signal processing unit 120. FIG. 3 is an explanatory diagram for describing detection of a target candidate by the target detection unit 130 according to the present embodiment.

  FIG. 3 shows a target O1 entering the detection area A1 of the radar 110. Note that the arrow in FIG. 3 indicates the traveling direction of the target O1. As shown in FIG. 3, when the target O1 enters the detection area A1 of the radar 110, an amplitude component depending on the material, the distance, and the area of the target O1 can be detected in the obtained frequency spectrum. At this time, the frequency spectrum is proportional to the speed of the target O1.

  The target detection unit 130 according to the present embodiment performs peak detection on the frequency spectrum in order to detect the amplitude component depending on the target O1. At this time, the target detecting unit 130 according to the present embodiment, for example, extracts only a plurality of amplitude components exceeding a predetermined threshold T and selects n pieces from the extracted plurality of amplitude components in ascending order, thereby performing peak detection. May go. As described above, in the present embodiment, the specific frequency that is the peak detection result of the frequency spectrum by the target detection unit 130 is referred to as a target candidate.

  FIG. 3 shows a reflection range R1 of the target O1 with respect to the radar 110. As described above, it can be seen that the portion of the target O1 where the radio wave is reflected depends on the shape of the target O1, the distance and the angle between the radar 110 and the target O1. For example, when the target O1 is a long target such as a truck, the distance between the two lines r1 and r2 in the width direction that constitutes the reflection range R1 is widened according to the vehicle length, and the target O1 is a light vehicle. In the case of a target having a short vehicle length as described above, the interval between the two lines r1 and r2 in the width direction forming the reflection range R1 becomes narrower according to the vehicle length.

  Here, assuming that the target O1 moves linearly at a constant speed, the speed of the target O1 observed by the radar 110 constitutes the moving direction of the target O1 and the reflection range R1. It depends on the angle between the two lines r1 and r2 in the width direction. That is, the frequency observed by the radar 110 has a bandwidth that depends on the vehicle length of the target O1. For this reason, it is expected that a feature relating to the vehicle length of the target O1 appears in the frequency bandwidth. Further, it is expected that the distribution tendency of the amplitude component with respect to the frequency includes a feature related to the shape (for example, unevenness) of the target O1. Further, by observing the temporal variation related to the above-described feature, an effect of more accurately capturing the property related to the shape of the target O1 is expected.

[2.4. Details of feature extraction]
Next, details of the feature amount extraction according to the present embodiment will be described. As described above, the feature extraction unit 140 according to the present embodiment extracts a feature amount related to a target candidate detected by the target detection unit 130. At this time, the feature extraction unit 140 according to the present embodiment may extract the above-described feature amount from a time-frequency expression (spectrogram) obtained by expanding the frequency spectrum of the target candidate in the time direction.

  FIG. 4 is a diagram illustrating an example of a data structure of a feature amount extracted by the feature extracting unit 140 according to the present embodiment. Here, in FIG. 4, the vertical axis represents frequency [Hz], and the horizontal axis represents time [s]. Each cell C shown in FIG. 4 corresponds to each time and frequency bin. FIG. 4 shows a target candidate OC1 detected by the target detection unit 130.

  As shown in FIG. 4, the feature extraction unit 140 according to the present embodiment can extract data (partial spectrogram) near the time and frequency of the target candidate OC1 as a feature amount. At this time, the feature extraction unit 140 according to the present embodiment, for example, falls within the range of −100 to +100 Hz from the frequency of the target target candidate OC1, and within one second from the time when the target candidate OC1 was obtained. Alternatively, all the amplitude value data to be extracted may be extracted as the feature amount. The feature extraction unit 140 may, for example, accumulate the data and select, as the feature amount, data whose time is earlier than the target candidate OC1. As described above, the feature extracting unit 140 locally extracts the feature amount from the frequency near the target candidate OC1, so that even when there are a plurality of targets having different speeds, the plurality of targets can be independently determined. Can be identified.

  Furthermore, when the radar 110 according to the present embodiment has a function of switching a plurality of oscillation frequencies in a stepwise manner and transmitting, the feature extraction unit 140 calculates the phase difference of the frequency spectrum between signals having different oscillation frequencies. Then, from the above equation (2), a spectrogram in which each cell C shown in FIG. 4 corresponds to a distance value can be obtained. In this case, the feature extracting unit 140 according to the present embodiment may select data (partial spectrogram) in the vicinity of the target target candidate OC1 from the spectrogram related to the distance and extract the data as a feature amount.

[2.5. Details by object sign]
Next, details of each object marker according to the present embodiment will be described. As described above, the object marker classification unit 160 according to the present embodiment has a function of performing object classification based on a machine learning model using the feature amount extracted by the feature extraction unit 140. At this time, the object marker classification unit 160 according to the present embodiment inputs one or both of a partial spectrogram related to the amplitude value and a partial spectrogram related to the distance value extracted by the feature extracting unit 140 as a feature amount. , The target can be identified.

  Further, one of the features of the object marker classification unit 160 according to the present embodiment is to realize highly accurate object marker classification by using analysis information on target candidates in addition to the above-described feature amounts. Here, the analysis information may include amplitude intensity, speed information, distance information, angle information, and the like related to the target candidate.

  For example, the landmark distinction unit 160 according to the present embodiment may identify a target using a multilayer neural network. FIG. 5 is a diagram illustrating a network structure example of the multilayer neural network according to the present embodiment.

  Referring to FIG. 5, the multilayer neural network according to the present embodiment includes a first input layer I1, a second input layer I2, a first layer L1, and a second layer L2.

  Here, the first input layer I1 may be an input layer for inputting the feature amount extracted by the feature extraction unit 140. As described above, the feature amount may include at least one of a partial spectrogram related to an amplitude value and a partial spectrogram related to a distance value extracted based on the target candidate. Also, referring to FIG. 5, the first input layer I1 is coupled to the first layer L1.

  Further, the first hierarchy L1 may be a convolutional neural network that abstracts the two-dimensional pattern shape of the feature quantity input from the first input layer I1. It is known that a convolutional neural network is compatible with data in which neighboring cells have meaning, such as the feature value according to the present embodiment. For this reason, in this embodiment, the effect of improving the target object identification performance is expected by employing the convolutional neural network.

  In addition, the first layer L1 according to the present embodiment may be configured by, for example, a plurality of Convolution layers, a Pooling layer, an activation function, and the like. Examples of the pooling layer include Max-Pooling and Sum-Pooling which are widely used. The activation function includes ReLU (Rectified Linear Unit), tanh, and the like. Note that the configuration of the first layer L1 according to the present embodiment is not limited to the above-described example, and can be flexibly modified. Also, referring to FIG. 5, the first layer L1 is connected to the second layer L2.

  Further, the second input layer I2 may be an input layer for inputting amplitude intensity, speed information, distance information, angle information, and the like relating to a target candidate obtained by the target detection unit 130. Referring to FIG. 5, the second input layer I2 is coupled to a second layer L2.

  The second layer L2 has a function of determining a target identification class based on target candidate analysis information input from the second input layer I2 and output from the first layer L2. Specifically, the second layer L2 according to the present embodiment includes a two-dimensional pattern of the feature amount abstracted by the first layer L1, the amplitude intensity, speed information, distance information, The identification class of the target may be output based on the angle information or the like.

  That is, in the multilayer neural network according to the present embodiment, in addition to learning the distribution pattern of the amplitude or the distance on the partial spectrogram, the distribution pattern can occur in any amplitude intensity, speed, distance, and angle-related situation. It is possible to learn in an integrated manner. For this reason, according to the multilayer neural network according to the present embodiment, the effect of learning the distribution pattern according to various situations can be expected.

  Note that the second layer L2 according to the present embodiment can be composed of, for example, one or more fully-connected layers and an activation function. The configuration of the second layer L2 according to the present embodiment can be flexibly modified, like the configuration of the first layer L1.

  As described above, the object marking unit 160 according to the present embodiment can determine the identification class of the object by using the multilayer neural network illustrated in FIG. In addition, the above-mentioned identification class may include a plurality of classes such as a small car, a medium-sized car, and a large car. Further, the identification class according to the present embodiment may include a class such as noise derived from hardware devices and a false image due to multiple reflection. In this case, even when a target other than the target is detected by the target detection unit 130 as a target candidate, the accuracy of identification can be increased.

[2.6. Target identification using tracking results]
Next, identification of a target using the tracking result according to the present embodiment will be described. The object marker classification unit 160 according to the present embodiment uses the tracking result of the target candidate by the target tracking unit 150 in addition to the identification class output by the multilayer neural network described above, so that a more accurate object can be obtained. It is possible to perform labeling.

  FIG. 6 illustrates an example of a tracking result output by the target tracking unit 150 according to the present embodiment. In an example shown in FIG. 6, a tracking result T1 relating to the target candidate OC is shown. At this time, the object marker classification unit 160 according to the present embodiment obtains the identification classes of the M target candidates OC related to the tracking result T1 by using the multilayer neural network described above, and determines the most frequently occurring identification class as the final object. It can be determined as a target identification class. In other words, the object marker classification unit 160 according to the present embodiment performs identification by a statistical method based on a set of target objects having a very high possibility of being the same target, thereby identifying a more accurate object marker. It is possible to realize.

Also, at this time, when the frequency difference f _diff in two or more different target candidates OC exceeds a predetermined threshold, the target marker classification unit 160 according to the present embodiment outputs the corresponding target output by the multilayer neural network. The identification class of the candidate OC may be excluded from the final identification class determination process. Specifically, the target marker classification unit 160 can exclude the above-described identification class corresponding to the target candidate OC that satisfies the criterion of the following Expression (4). Note that f _th in the following equation (4) may be any positive real number.

  Even when the extracted partial spectrogram includes amplitude components derived from two or more target candidates, the target marker discriminating unit 160 according to the present embodiment performs the processing based on the above equation (4). , It is possible to exclude the identification result related to the partial spectrogram.

[2.7. Process flow of information processing device 10]
Next, a processing flow of the information processing apparatus 10 according to the present embodiment will be described. FIG. 7 is a flowchart illustrating a flow of a process of the information processing apparatus 10 according to the present embodiment.

  Referring to FIG. 7, first, the signal processing unit 120 of the information processing device 10 performs frequency analysis on a signal received by the radar 110, and calculates a frequency spectrum related to a Doppler frequency (S1101).

  Next, the target detection unit 130 detects at least one or more target candidates based on the intensity of the frequency spectrum acquired in step S1101 (S1102).

  Next, the feature extraction unit 140 extracts a feature amount related to the target candidate detected in step S1102 (1103).

  Next, the target tracking unit 150 performs tracking of the target candidate detected in step S1102, and outputs a tracking result (S1104).

  Next, the object marker classification unit 160 performs object marker classification based on a machine learning model using the feature amount extracted in step S1103 and the analysis information of the target candidate detected in step S1102 (S1105). .

  Subsequently, based on the identification model output in step S1105 and the tracking result output in step S1104, the target marker classification unit 160 determines the final identification class of the target, and outputs the identification result. (S1106).

<3. Example of hardware configuration>
Next, an example of a hardware configuration of the information processing apparatus 10 according to the present invention will be described. FIG. 8 is a block diagram illustrating a hardware configuration example of the information processing apparatus 10 according to the present invention. Referring to FIG. 8, for example, the information processing apparatus 10 includes a CPU 871, a ROM 872, a RAM 873, a host bus 874, a bridge 875, an external bus 876, an interface 877, an input unit 878, and an output unit 879. , A storage unit 880, a drive 881, a connection port 882, and a communication unit 883. Note that the hardware configuration shown here is an example, and some of the components may be omitted. Further, components other than the components shown here may be further included.

(CPU 871)
The CPU 871 functions as, for example, an arithmetic processing device or a control device, and controls the entire operation of each component or a part thereof based on various programs recorded in the ROM 872, the RAM 873, the storage unit 880, or the removable recording medium 901. .

(ROM 872, RAM 873)
The ROM 872 is a means for storing programs read by the CPU 871, data used for calculations, and the like. The RAM 873 temporarily or permanently stores, for example, a program read by the CPU 871 and various parameters that appropriately change when the program is executed.

(Host bus 874, bridge 875, external bus 876, interface 877)
The CPU 871, the ROM 872, and the RAM 873 are connected to each other via, for example, a host bus 874 capable of high-speed data transmission. On the other hand, the host bus 874 is connected to, for example, an external bus 876 having a relatively low data transmission speed via a bridge 875. The external bus 876 is connected to various components via an interface 877.

(Input unit 878)
For the input unit 878, for example, a mouse, a keyboard, a touch panel, a button, a switch, a microphone, a lever, and the like are used. Further, as the input unit 878, a remote controller (hereinafter, remote controller) capable of transmitting a control signal using infrared rays or other radio waves may be used.

(Output unit 879)
The output unit 879 includes, for example, a display device (display device) such as a CRT (Cathode Ray Tube), an LCD, or an organic EL, an audio output device such as a speaker or a headphone, a printer, a mobile phone, or a facsimile. Is a device capable of visually or audibly notifying a user. In addition, the output unit 879 may include various radars as described above.

(Storage unit 880)
The storage unit 880 is a device for storing various data. As the storage unit 880, for example, a magnetic storage device such as a hard disk drive (HDD), a semiconductor storage device, an optical storage device, or a magneto-optical storage device is used.

(Drive 881)
The drive 881 is, for example, a device that reads information recorded on a removable recording medium 901 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, or writes information on the removable recording medium 901.

(Removable recording medium 901)
The removable recording medium 901 is, for example, a DVD medium, a Blu-ray (registered trademark) medium, an HD DVD medium, various semiconductor storage media, and the like. Of course, the removable recording medium 901 may be, for example, an IC card on which a non-contact type IC chip is mounted, or an electronic device.

(Connection port 882)
The connection port 882 is, for example, a port for connecting an external connection device 902 such as a USB (Universal Serial Bus) port, an IEEE1394 port, a SCSI (Small Computer System Interface), an RS-232C port, or an optical audio terminal. is there.

(External connection device 902)
The external connection device 902 is, for example, a printer, a portable music player, a digital camera, a digital video camera, or an IC recorder.

(Communication unit 883)
The communication unit 883 is a communication device for connecting to the network 903. For example, a communication card for a wired or wireless LAN, Bluetooth (registered trademark), or WUSB (Wireless USB), a router for optical communication, an ADSL (Asymmetric) Digital Subscriber Line) or a modem for various communications. Further, it may be connected to a telephone network such as an extension telephone network or a mobile telephone carrier network.

<4. Summary>
As described above, one of the features of the information processing apparatus 10 according to the present invention is that the information processing apparatus 10 performs object labeling using a multilayer neural network. Further, in the multilayer neural network according to the present invention, in addition to learning the distribution pattern of the amplitude or the distance on the partial spectrogram, what kind of amplitude intensity, speed, distance, and angle can be generated in the distribution pattern? Can be integratedly learned. According to such a configuration, it is possible to realize highly accurate object labeling at a lower cost.

  As described above, the preferred embodiments of the present invention have been described in detail with reference to the accompanying drawings, but the present invention is not limited to such examples. It is obvious that those skilled in the art to which the present invention pertains can conceive various changes or modifications within the scope of the technical idea described in the claims. It is understood that these also belong to the technical scope of the present invention.

  For example, in the above embodiment, the vehicle has been described as an example of the target, but the present invention is not limited to this example. For example, the target according to the present embodiment may include a ship, an unmanned aerial vehicle, and the like.

  Further, in the above embodiment, a radar that radiates radio waves has been described as an example, but the present invention may be implemented using a radar that radiates sound waves.

  Further, in the above embodiment, the case where the amplitude value of the spectrogram by the fast Fourier transform is used has been described as an example. However, in the present invention, the wavelet coefficient of the spectrogram by the wavelet transform may be used.

  In the above-described embodiment, the case where the target tracking unit 150 performs tracking based on the speed of the target candidate has been described as an example. However, tracking of the target candidate according to the present invention is based on the distance of the target candidate. Or may be realized based on both the speed and the distance of the target candidate.

  Further, in the above embodiment, the case where the first layer L1 and the second layer L2 are directly connected is described as an example, but the network structure of the multilayer neural network according to the present invention is not limited to such an example. It goes without saying.

  Further, each step in the processing of the information processing apparatus 10 of the present invention does not necessarily need to be processed in a time series in the order described in the flowchart. For example, each step in the processing of the information processing device 10 may be processed in an order different from the order described in the flowchart, or may be processed in parallel.

REFERENCE SIGNS LIST 10 information processing device 110 radar 120 signal processing unit 130 target detection unit 140 feature extraction unit 150 target tracking unit 160 target sign distinction unit I1 first input layer I2 second input layer L1 first layer L2 second hierarchy

Claims (9)

A first hierarchy that abstracts a two-dimensional pattern shape of a feature quantity extracted based on a frequency spectrum of the detected target, and analysis information of target candidates detected from the frequency spectrum of the target And a second layer that determines the identification class of the target based on the output from the first layer and a second layer that determines the classification of the target based on the output from the first layer.
With
The feature amount includes at least one of a partial spectrogram related to an amplitude value or a partial spectrogram related to a distance value extracted from a spectrogram based on the frequency spectrum,
Information processing device. The analysis information of the target candidate includes at least one of amplitude intensity, speed information, distance information, or angle information,
The information processing device according to claim 1. A feature extraction unit that extracts data near the target candidate from the spectrogram obtained by expanding the frequency spectrum in the time direction as the feature amount;
Further comprising,
The information processing apparatus according to claim 1. A target detection unit that detects at least one or more target candidates based on the intensity of the frequency spectrum;
Further comprising,
The information processing device according to claim 1. A target tracking unit that outputs a tracking result indicating a temporal correspondence relationship between the plurality of target candidates,
Further comprising
The target marker distinction unit performs the identification of the target based on the identification class and the tracking result output by the multilayer neural network,
The information processing apparatus according to claim 1. The target marker distinction unit performs the identification of the target based on the most frequently occurring identification class among the plurality of identification classes corresponding to the tracking result,
The information processing device according to claim 5. When the frequency difference between two or more different target candidates exceeds a predetermined threshold, the target marker separate unit does not use the identification class related to the target candidate for the identification of the target.
The information processing device according to claim 6. Computer
A first hierarchy that abstracts a two-dimensional pattern shape of a feature quantity extracted based on a frequency spectrum of the detected target, and analysis information of target candidates detected from the frequency spectrum of the target And a second layer that determines the identification class of the target based on the output from the first layer and a second layer that determines the classification of the target based on the output from the first layer.
With
The feature amount includes at least one of a partial spectrogram related to an amplitude value or a partial spectrogram related to a distance value extracted from a spectrogram based on the frequency spectrum,
Information processing equipment,
Program to function as An information processing method using a multilayer neural network,
Abstracting the two-dimensional pattern shape of the feature quantity extracted based on the frequency spectrum of the detected target;
Determining the identification class of the target based on the analysis information of the target candidate detected from the frequency spectrum of the target and the two-dimensional pattern of the abstracted feature amount;
Including
The feature amount includes at least one of a partial spectrogram related to an amplitude value or a partial spectrogram related to a distance value extracted from a spectrogram based on the frequency spectrum,
Information processing method.

Images