阅读说明：本技术 雷达装置 (Radar apparatus ) 是由 清水直继 藤津圣也 于 2020-04-09 设计创作，主要内容包括：搭载于移动体的雷达装置(10)的特征量计算部(S410)使用与提取出的各峰值建立对应关系的信息,计算预先决定的一种以上的特征量(D1～D10)。环境判定部(S420～S460)根据特征量计算部中的计算结果,使用表示是特定环境的概率的肯定分布计算特定环境概率(P(Di|P)),并且使用表示是非特定环境的概率的否定分布计算非特定环境概率(P(Di|N))。环境判定部进一步根据统合按照每个特征量计算的特定环境概率和非特定环境概率而得到的结果,判定移动体(车辆)处于特定环境下还是处于非特定环境下。根据本发明,能够在搭载于移动体的雷达装置中,提高移动体是否处于特定环境下的判定精度。(A feature value calculation unit (S410) of a radar device (10) mounted on a mobile body calculates one or more predetermined feature values (D1-D10) using information that correlates with each extracted peak value. The environment determination unit (S420-S460) calculates the specific environment probability (P (Di | P)) using a positive distribution indicating the probability of being a specific environment and calculates the non-specific environment probability (P (Di | N)) using a negative distribution indicating the probability of being a non-specific environment, based on the calculation result in the feature amount calculation unit. The environment determination unit further determines whether the moving object (vehicle) is in the specific environment or the non-specific environment based on a result of integrating the specific environment probability and the non-specific environment probability calculated for each feature amount. According to the present invention, in a radar device mounted on a moving object, it is possible to improve the accuracy of determining whether or not the moving object is in a specific environment.)

1. A radar device mounted on a mobile body, the radar device comprising:

a spectrum generation unit (70: S110 to S160) configured to generate at least one of a one-dimensional spectrum and a two-dimensional spectrum by analyzing a signal obtained by transmitting and receiving a modulated wave based on each modulation method as a radar wave using one or more modulation methods;

a peak extraction unit (70: S170) configured to extract a peak from at least one of the one-dimensional spectrum and the two-dimensional spectrum generated for each of the modulation methods;

a feature value calculation unit (70: S310, S410) configured to calculate one or more predetermined feature values using information associated with the respective peak values extracted by the peak value extraction unit;

a distribution storage unit (72) that stores in advance a positive distribution and a negative distribution, the positive distribution being generated in advance for each of the feature amounts, the positive distribution being a distribution that indicates a probability that, when the feature amount is given, an environment in which the feature amount is acquired is a predetermined specific environment, and the negative distribution being a distribution that indicates a probability that, when the feature amount is given, an environment in which the feature amount is acquired is a non-specific environment other than the specific environment; and

an environment determination unit (70: S420 to S460) configured to calculate a specific environment probability using the positive distribution and a non-specific environment probability using the negative distribution based on the calculation result of the feature amount calculation unit, determine whether the mobile object is in the specific environment or the non-specific environment based on a result obtained by integrating the specific environment probability and the non-specific environment probability calculated for each feature amount,

the distribution shapes are different between the case where the feature amount is calculated in the specific environment and the case where the feature amount is calculated in the non-specific environment.

2. The radar apparatus of claim 1,

the feature value calculating unit calculates an average azimuth of the peak value extracted by the peak value extracting unit as one of the feature values.

3. The radar apparatus according to claim 1 or 2,

the feature value calculation unit calculates a variance of the azimuth of the peak extracted by the peak extraction unit as one of the feature values.

4. The radar apparatus according to claim 2 or 3,

the feature amount calculation unit uses the azimuth difference from a preset reference direction as the azimuth of the peak extracted by the peak extraction unit.

5. The radar apparatus according to any one of claims 1 to 4,

the feature amount calculation unit calculates at least one of an average power of the plurality of peaks extracted by the peak extraction unit, an average power ratio of the plurality of peaks to the noise floor, a variance of powers of the plurality of peaks, and a variance of power ratios of the plurality of peaks to the noise floor as one of the feature amounts.

6. The radar apparatus according to any one of claims 1 to 5,

the feature amount calculation unit calculates a total number of peaks extracted by the peak extraction unit as one of the feature amounts.

7. The radar apparatus according to any one of claims 1 to 6,

the spectrum generation unit is configured to: an FFT spectrum that is the one-dimensional spectrum is generated by performing FFT on a signal obtained by at least one of the above modulation schemes, and the two-dimensional spectrum is generated by expanding the FFT spectrum azimuth by each frequency bin.

8. The radar apparatus of claim 7,

the feature amount calculation unit calculates at least one of an average power of a plurality of peaks detected on the FFT spectrum, a power ratio of the plurality of peaks to a noise floor, a power variance of the plurality of peaks, and a power variance of the plurality of peaks and the noise floor as one of the feature amounts.

9. The radar apparatus according to claim 7 or 8,

the feature amount calculation unit calculates at least one of an average power of the FFT spectrum, a power ratio of the FFT spectrum to a noise floor, a power variance of the FFT spectrum, and a variance of the power ratio of the FFT spectrum to the noise floor as one of the feature amounts.

10. The radar apparatus according to any one of claims 7 to 9,

the feature amount calculation unit calculates a total number of peaks extracted in the FFT spectrum as one of the feature amounts.

11. The radar apparatus according to any one of claims 8 to 10,

when the peak azimuth on the FFT spectrum is expanded, the feature amount calculation unit calculates an average of the number of peaks separated from one peak on the FFT spectrum on the two-dimensional spectrum as one of the feature amounts.

12. The radar apparatus according to any one of claims 1 to 11,

the feature amount calculation unit calculates the feature amount using a peak value detected at a distance equal to or greater than a threshold value set according to a moving speed of the mobile body.

13. The radar apparatus according to any one of claims 1 to 12,

the above modulation schemes include at least FMCW modulation,

the spectrum generation unit generates the one-dimensional spectrum and the two-dimensional spectrum for at least one of the modulation schemes using both the uplink chirp and the downlink chirp in the FMCW modulation.

14. The radar apparatus according to any one of claims 1 to 13,

the modulation method at least comprises multi-frequency CW modulation or FCM modulation.

15. The radar apparatus according to any one of claims 1 to 14, further comprising:

a collection unit (70: S310 to S320) that collects the feature amount calculated by the feature amount calculation unit in association with teacher data indicating whether the environment in which the feature amount was acquired is the specific environment or the unspecified environment;

and a learning unit (70: S330 to S350) that generates the positive distribution and the negative distribution using the feature values collected by the collection unit, based on a preset update timing, and updates the storage content of the distribution storage unit.

Technical Field

The present disclosure relates to a radar device mounted on a mobile body.

Background

A radar device mounted on a moving object such as a vehicle is likely to cause erroneous detection of a target object when used in an environment where a very large number of reflection points are detected, such as a tunnel and a stereo parking lot.

Patent document 1 describes the following technique: in a radar device, it is determined whether or not a vehicle is in a specific environment in which erroneous detection is likely to occur, based on a variation in the number of detected peaks, and when it is determined that the vehicle is in the specific environment, erroneous detection is suppressed by raising a threshold value at the time of extracting the detected peaks.

Patent document 1: japanese patent laid-open publication No. 2016-206011

However, the inventors have found the following problems as a result of their detailed studies.

In a specific environment such as a tunnel surrounded by a wall surface, not only the number of reflection points is increased, but also the spread of radio waves is suppressed, and therefore, reflected power from a target object existing at a long distance is detected at a power higher than that outside the specific environment. In particular, in a method of calculating a distance using a phase difference of a signal, in other words, a phase of a difference signal, such as double-frequency CW or FCM, the calculated distance includes ambiguity due to phase folding. In other words, the distance in which the phase of the signal rotates once is the detection upper limit distance, and a target object located at a position farther than the detection upper limit distance is erroneously detected as a target object within the detection upper limit distance due to the phase return. Such ghosts cannot be removed only by changing the threshold for peak detection, and a special process for removing ghosts, that is, a ghosting removal process, needs to be additionally performed. However, the ghost elimination processing is preferably performed only when necessary because it is necessary to avoid unnecessary deletion of the detection target due to erroneous determination of ghost and the processing load on the radar device increases.

Disclosure of Invention

In one aspect of the present disclosure, a technique for improving the accuracy of determination as to whether or not a moving object is in a specific environment in a radar device mounted on the moving object may be provided.

One aspect of the present disclosure is a radar device mounted on a mobile body, including: the spectrum generation unit, the peak extraction unit, the feature calculation unit, the distribution storage unit, and the environment determination unit.

The spectrum generation unit generates at least one of a one-dimensional spectrum and a two-dimensional spectrum by analyzing a signal obtained by transmitting and receiving a modulated wave based on each modulation method as a radar wave using one or more modulation methods. The peak extraction unit extracts a peak from at least one of the one-dimensional spectrum and the two-dimensional spectrum generated for each modulation scheme. The feature value calculating unit calculates one or more predetermined feature values using information associated with the respective peak values extracted by the peak value extracting unit. The distribution storage unit stores in advance a positive distribution and a negative distribution, the positive distribution being generated in advance for each feature amount, the positive distribution indicating a probability that an environment when the feature amount is acquired is a predetermined specific environment when the feature amount is given, and the negative distribution indicating a probability that the environment when the feature amount is acquired is a non-specific environment other than the specific environment when the feature amount is given. The environment determination section calculates a specific environment probability using a positive distribution and a non-specific environment probability using a negative distribution, according to a calculation result in the feature amount calculation section. The environment determination unit further determines whether the mobile object is in the specific environment or the non-specific environment based on a result of integrating the specific environment probability and the non-specific environment probability calculated for each feature amount. Wherein different distribution shapes are followed in the case where the feature amount is calculated under a specific environment and in the case where the feature amount is calculated under a non-specific environment.

According to such a configuration, since it is possible to perform the determination of whether or not the specific environment is present by combining a plurality of feature amounts, it is possible to improve the determination accuracy. In addition, the physical quantity that corresponds to the coordinate axes of the one-dimensional spectrum and the two-dimensional spectrum differs depending on the modulation method. For example, in 2FCW, the coordinate axis of the one-dimensional spectrum is associated with a relative velocity, and in FMCW, the coordinate axis of the one-dimensional spectrum is associated with a distance and a relative velocity. For example, in 2FCW, the coordinate axis of the two-dimensional spectrum corresponds to the relative velocity and the orientation, in FMCW, the coordinate axis of the two-dimensional spectrum corresponds to the distance and the relative velocity and the orientation, and in FCM, the coordinate axis of the two-dimensional spectrum corresponds to the distance and the relative velocity. It should be noted that FMCW is an abbreviation for Frequency Modulated Continuous Wave, CW is an abbreviation for Continuous Wave, and FCM is an abbreviation for Fast-Chirp Modulation. 2F of 2FCW means two frequencies.

Drawings

Fig. 1 is a block diagram showing the configuration of an in-vehicle system.

Fig. 2 is an explanatory diagram of a modulation scheme of a radar wave used.

Fig. 3 is a flowchart of the target detection process.

Fig. 4 is an explanatory diagram illustrating an FFT spectrum generated by the target object detection processing and a two-dimensional spectrum after azimuth expansion.

Fig. 5 is an explanatory diagram illustrating an FFT spectrum and a determination region in a two-dimensional spectrum.

Fig. 6 is a histogram of the first feature amount and a normal distribution.

Fig. 7 is a histogram of the second feature amount and a normal distribution.

Fig. 8 is a histogram of the fourth feature quantity, the fifth feature quantity, and the feature quantity similar to the fourth feature quantity.

Fig. 9 is a histogram of the sixth feature amount.

Fig. 10 is a histogram of the seventh feature amount and the eighth feature amount.

Fig. 11 is a histogram of the ninth feature amount and the tenth feature amount.

Fig. 12 is a flowchart of the environment determination processing in the first embodiment.

Fig. 13 is a flowchart of the environment determination processing in the second embodiment.

Fig. 14 is a graph in which the integrated probability calculated when the vehicle travels in the non-specific environment and the logarithmic bayesian ratio before and after the filter processing are plotted for each processing cycle.

Fig. 15 is a graph plotting integration probabilities calculated while traveling in a specific environment and logarithmic bayesian ratios before and after the filter processing for each processing cycle.

Fig. 16 is a flowchart of the target detection process in the third embodiment.

Fig. 17 is a flowchart of the learning process.

Detailed Description

Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.

[0. specific Environment ]

First, a specific environment referred to in the present disclosure is explained.

In a straight tunnel, since the reflections from the ceiling or the roadside object are aligned behind the vehicle, the direction of the ceiling or the roadside object detected by the radar matches the direction of detection of the vehicle approaching in the tunnel. In addition, in the tunnel, no radio wave is diffused, and the reflected power of the object existing at a long distance is detected at a power higher than that outside the tunnel. Therefore, in a modulation method in which the calculation of the distance uses the phase of the radio wave (that is, the calculation of the distance is ambiguous), an object existing at a distance farther than the detection distance rotated by one rotation of the phase is detected within the detection distance, and a so-called ghost occurs. Hereinafter, such an illusion is referred to as a long-range illusion.

In addition, in the case of a radar that detects a target object by comparing information such as a peak detected by each modulation with a plurality of modulation methods, a long-distance phantom detected by a certain modulation method and a reflected wave from a ceiling in a short distance detected by another modulation method are detected at the same distance and direction, and these are mistaken for a reflection point based on the same target object. Therefore, the erroneous determination is that the target object is present at a short distance, which may cause malfunction such as an alarm. In the present disclosure, at least an environment in which there is a possibility that a long-distance ghost is erroneously detected as a short-distance object is set as a specific environment. However, the specific environment that can be handled in the present disclosure is not limited thereto.

[1. first embodiment ]

[ 1-1. Structure ]

As shown in fig. 1, the in-vehicle system 1 includes a radar device 10 and a driving assistance ECU 100. The ECU is an abbreviation of Electronic Control Unit (Electronic Control Unit). The in-vehicle system 1 is mounted on a vehicle such as a four-wheel automobile as a moving body. The radar devices 10 are attached to, for example, the rear end and the left and right ends of the vehicle, respectively, and are arranged so as to include the rear along the traveling direction of the vehicle and the lateral direction orthogonal to the traveling direction within the detection range of the radar devices 10. The vehicle on which the radar device 10 is mounted is also referred to as a host vehicle.

The radar device 10 emits a radar wave and receives a reflected wave, and observes a distance R to a target object, a speed V of the target object, and an azimuth θ of the target object based on the received signal Sr. The radar device 10 calculates estimated values of the lateral position x, the longitudinal position y, the lateral velocity Vx, and the longitudinal velocity Vy from these observed values (R, V, θ), and inputs these estimated values (x, y, Vx, Vy) to the driving assistance ECU 100. The lateral position x is a position along the vehicle width direction of the vehicle on which the in-vehicle system 1 is mounted, and the longitudinal position y is a position along the traveling direction of the vehicle.

The driving assistance ECU100 executes various processes for assisting the driver in driving the vehicle based on the estimated values (x, y, Vx, Vy) of the respective target objects input from the radar device 10. The processing related to the driving assistance may include, for example, processing of giving a warning to the driver that an approaching object exists, processing of executing vehicle control for avoiding a collision with an approaching object by controlling a brake system, a steering system, or the like, automatically changing lanes, or the like.

The radar device 10 includes a transmission circuit 20, a distributor 30, a transmission antenna 40, a reception antenna 50, a reception circuit 60, a processing unit 70, and an output unit 80.

The transmission circuit 20 is a circuit for supplying the transmission signal Ss to the transmission antenna 40. The transmission circuit 20 inputs a high-frequency signal in a millimeter-wave band to the distributor 30 located upstream of the transmission antenna 40. Specifically, as shown in fig. 2, the transmission circuit 20 alternately repeats the first modulation period and the second modulation period for each predetermined processing cycle, and inputs the high-frequency signal generated in each modulation period to the distributor 30. Further, during the first modulation, a high-frequency signal is generated, which is frequency-modulated so as to increase in frequency and decrease in frequency into a triangular wave shape. During the second modulation, a high-frequency signal whose frequency is alternately switched is generated. The processing cycle is set to be longer than the total period of the first modulation period and the second modulation period, and a period before the first modulation period after the second modulation period ends and the next processing cycle starts is referred to as a processing period.

In other words, the radar device 10 operates as an FMCW radar that transmits and receives FMCW as a first modulated wave in the first modulation period, and operates as a 2FCW radar that transmits a double-frequency CW (hereinafter, 2FCW) as a second modulated wave in the second modulation period. The two frequencies used in the 2FCW are set so that the distance can be uniquely measured within a range of a predetermined upper limit distance (for example, 150 m). Hereinafter, two signals having different frequencies used in 2FCW will be referred to as a first signal and a second signal. The waveform of the FMCW is set so that the distance can be uniquely determined within a distance range different from the predetermined upper limit distance. Further, the upper limit distance based on the first modulation and the upper limit distance based on the second modulation may be the same distance.

Returning to fig. 1, the distributor 30 distributes the high-frequency signal power input from the transmission circuit 20 into the transmission signal Ss and the local signal L.

The transmitting antenna 40 emits a radar wave of a frequency corresponding to the transmission signal Ss based on the transmission signal Ss supplied from the distributor 30.

The receiving antenna 50 is an antenna for receiving a reflected wave, which is a radar wave reflected by a target object. The receiving antenna 50 is configured as a linear array antenna in which a plurality of antenna elements 51 are arranged in a row. The reception signal Sr of the reflected wave of each antenna element 51 is input to the reception circuit 60.

The reception circuit 60 processes the reception signal Sr input from each antenna element 51 constituting the reception antenna 50, and generates and outputs a beat signal BT for each antenna element 51. Specifically, the reception circuit 60 mixes the reception signal Sr input from the antenna element 51 and the local signal L input from the distributor 30 using the mixer 61 for each antenna element 51, thereby generating and outputting the beat signal BT for each antenna element 51.

Wherein the process until the beat signal BT is output includes: a process of amplifying the reception signal Sr, a process of removing unnecessary signal components from the beat signal BT, and a process of converting the beat signal BT into digital data. In this way, the reception circuit 60 converts the generated beat signal BT for each antenna element 51 into digital data and outputs it. The output beat signal BT for each antenna element 51 is input to the processing unit 70. Hereinafter, the a/D conversion data of the beat signal BT acquired during the first modulation period is referred to as first modulation data, and the a/D conversion data of the beat signal BT acquired during the second modulation period is referred to as second modulation data.

The processing unit 70 includes a microcomputer having a CPU71 and a semiconductor memory (hereinafter, memory 72) such as a RAM or a ROM. The processing unit 70 may be provided with a coprocessor that performs fast fourier transform (hereinafter, FFT) processing and the like.

The processing unit 70 performs at least target detection processing. The target detection processing is processing for analyzing the beat signal BT for each antenna element 51 to calculate an estimated value (x, y, Vx, Vy) for each target of the reflected radar wave.

[ 1-2. treatment ]

[ 1-2-1. target detection treatment ]

The target detection process executed by the processing unit 70 will be described with reference to the flowchart of fig. 3.

When the in-vehicle system 1 is started, the present process is repeatedly executed for each process cycle.

When the present process is started, the processing unit 70 determines in S110 whether the first modulation period is finished, that is, whether the acquisition of the first modulated data is finished. If the acquisition of the first modulated data is not completed, the processing unit 70 waits by repeating this step, and if the acquisition of the first modulated data is completed, the process proceeds to S120.

In S120, the processing unit 70 calculates a power spectrum by performing frequency analysis processing on the first modulated data for each antenna element 51 and for each of the uplink chirp and the downlink chirp.

Here, the FFT processing is performed as frequency resolution processing. FFT is an abbreviation for Fast Fourier Transform. The power spectrum obtained by the FFT is referred to as an FFT spectrum. In the FFT spectrum, the power of the reflected wave is represented by each frequency bin. The frequency bin is a frequency range that is a unit scale of the FFT spectrum, and is determined by the number of samples of data to be FFT and the sampling frequency.

In addition, in FMCW, the uplink chirp is a signal whose frequency increases with time, and in FMCW, the downlink chirp is a signal whose frequency decreases with time. Hereinafter, the FFT spectrum of the uplink chirp is referred to as an UP-FFT spectrum, and the FFT spectrum of the downlink chirp is referred to as a DN-FFT spectrum. These UP-FFT spectrum and DN-FFT spectrum correspond to a one-dimensional spectrum.

Processing section 70 calculates an average FFT spectrum obtained by averaging FFT spectra obtained from each antenna element 51 for each of the UP-FFT spectrum and the DN-FFT spectrum. Then, processing section 70 extracts a frequency bin having a peak whose signal level is equal to or higher than a predetermined threshold on the average FFT spectrum.

In the following S130, the processing unit 70 performs an azimuth operation on the UP-FFT spectrum and the DN-FFT spectrum calculated in S120, respectively.

In the azimuth calculation, azimuth expansion is performed using the fact that the phases of peaks detected in the same frequency bin of each channel are different for each channel. By this azimuth calculation, a two-dimensional spectrum is generated with the frequency bin and the azimuth as coordinate axes. High resolution algorithms such as MUSIC can also be used in the azimuth calculation. MUSIC is an abbreviation for Multiple Signal Classification. Not limited to this, beam forming or the like may be used. In addition, the azimuth operation is performed at least on all the frequency bins where the peak is detected on the FFT spectrum in the previous S120. Hereinafter, the two-dimensional spectrum of the uplink chirp is referred to as an UP spectrum, and the two-dimensional spectrum of the downlink chirp is referred to as a DN spectrum.

In the next S140, the processing unit 70 determines whether the second modulation period is finished, that is, whether the acquisition of the second modulated data is finished. If the second modulation period is not completed, the processing unit 70 waits by repeating this step, and if the second modulation period is completed, the processing unit 70 shifts the process to S150.

In S150, the processing unit 70 performs frequency analysis processing on the second modulated data per antenna element 51 and for each of the first signal and the second signal to generate a power spectrum, and detects a peak on the power spectrum. Here, as in S120, FFT processing is used as the frequency analysis processing. The FFT spectrum obtained as a result of this FFT processing also corresponds to a one-dimensional spectrum.

In addition, since the frequencies of the first signal and the second signal in the 2FCW are sufficiently close to each other, the doppler frequency detected from the first signal and the doppler frequency detected from the second signal have substantially the same magnitude. In other words, in the FFT spectrum of the first signal and the FFT spectrum of the second signal, peaks are detected in the same frequency bin. As a result, the FFT spectrum of the first signal and the FFT spectrum of the second signal have the same shape, and therefore only one FFT spectrum is shown in fig. 4.

Then, an average FFT spectrum obtained by averaging the FFT spectra obtained for each antenna element 51 is calculated for each of the first signal and the second signal, and a frequency bin having a peak value with a power equal to or higher than a predetermined threshold value is extracted.

Further, the distance is calculated from the phase difference Δ θ of two peak frequency components detected in the same frequency bin from the two average FFT spectra. However, since the actual phase difference cannot be distinguished as Δ θ or 2n π + Δ θ, the distance calculated from the phase difference Δ θ is ambiguous. N is an integer.

In next S160, processing section 70 performs the azimuth calculation using the FFT spectrum (hereinafter, MF-FFT spectrum) of either one of the first signal and the second signal, in the same manner as in S130. The generated two-dimensional spectrum is referred to as an MF spectrum by the azimuth calculation. The azimuth operation is performed at least for all frequency bins where a peak was detected on the MF-FFT spectrum in the previous S150.

In next S170, processing section 70 extracts, as target peaks, all peaks having power equal to or higher than a predetermined threshold from the UP spectrum and DN spectrum generated in S130 and the MF spectrum generated in S160.

In next S180, the processing unit 70 performs peak matching for estimating that peaks based on the same target object are associated with each other among the target peaks extracted in S170, and generates instantaneous values in the period.

Specifically, an estimation range of a frequency bin in which a target peak appears in the UP spectrum and the DN spectrum is set based on the azimuth, the distance, and the relative velocity determined from the target peak on the MF spectrum, and a peak existing in the estimation range is extracted.

As shown in fig. 4, the estimation ranges set here are all set to the same orientation as the target peak, and the range of the frequency bin can be variably set according to the distance and the relative speed. Specifically, when the reflection point corresponding to the target peak is approaching the host vehicle, the range of the frequency bin is set so that the UP spectrum is lower than the DN spectrum. In the case where the reflection point corresponding to the target peak is moving away from the own vehicle, the range of the frequency bin is set so that the UP spectrum is higher than the DN spectrum. This is based on the characteristic of the FMCW in which the magnitude relationship between the frequency detected in the up-chirp and the frequency detected in the down-chirp varies due to the doppler shift. However, when the relative speed between the reflection point corresponding to the target peak and the host vehicle is zero, the peak in 2FCW is hidden in low-frequency noise and cannot be detected, and therefore cannot be matched with the peak in FMCW. In this case, matching of the peak of the UP spectrum and the peak of the DN spectrum is performed only in FMCW.

Hereinafter, a peak on the XX spectrum is referred to as an XX peak. Wherein XX represents any one of MF-FFT, UP-FFT, DN-FFT, MF, UP, and DN.

Then, (MF, UP, DN) represents an instantaneous value when both UP and DN peaks corresponding to the MF peak are extracted. An instantaneous value in the case where only the UP peak corresponding to the MF peak is extracted is represented by (MF, UP). An instantaneous value in the case where only a DN peak corresponding to an MF peak is extracted is represented by (MF, DN).

An instantaneous value extracted by a method known in FMCW radar using an UP peak and a DN peak, instead of using an MF peak, is represented by (UP, DN).

In next S190, the processing unit 70 performs tracking using the instantaneous value generated in S180 to detect the target object. In the tracking, the distance and the direction (hereinafter, predicted position) at which the target object or the like is predicted to be detected in the current processing cycle are calculated from the target object detected in the previous processing cycle and the target object candidate (hereinafter, target object or the like). Further, the distance and the orientation (hereinafter, detection position) of the reflection point (hereinafter, peak value corresponding point) represented by the instantaneous value are calculated from the instantaneous value. When the difference between the predicted position and the detected position is within a preset allowable range, the target object and the instantaneous value are correlated with the reflection from the same object, and history is performed. If the correspondence relationship cannot be established with any target object, the instantaneous value is determined as a newly detected instantaneous value without a history connection, and the instantaneous value is determined as a new target object candidate. When the history connection is confirmed after a predetermined number of processing cycles, the target object candidate is recognized as a true target object.

In the next S200, the processing unit 70 executes an environment determination process of determining whether or not the surrounding situation in which the host vehicle is traveling is a specific environment. The details of the environment determination process will be described later.

In next S210, if it is determined that the result of the environment determination processing in S200 is the specific environment, the processing unit 70 moves the processing to S220, and if it is determined that the result is not the specific environment, the processing unit 70 moves the processing to S230.

In S220, processing section 70 determines whether or not each target object or the like is a virtual image, sets a virtual image flag according to the determination result, and advances the process to S230. The determination as to whether or not the target object or the like is a virtual image may be made, for example, by extracting one or more feature amounts from information associated with the target object or the like and instantaneous values that have a historical connection with the target object or the like, and probabilistically determining the feature amounts using the same method as the environment determination processing.

In S240, the processing unit 70 generates the estimated values (x, y, Vx, Vy) relating to the target object detected in S190, outputs the values to the driving assistance ECU100 via the output unit 80, and ends the processing. The inferred value may also include the virtual image flag F.

In addition, in the driving assistance ECU100, when the estimated value includes the virtual flag F, for example, the virtual flag is used as the warning on/off information, and the estimated value (that is, the target object) with respect to which the virtual flag F is set on may be excluded from objects that control the warning, the brake system, the steering system, and the like. The virtual image flag F may be used for various other controls.

In the target detection process, S110 to S160 correspond to a spectrum generation unit, and S170 corresponds to a peak extraction unit.

[ 1-2-2. characteristic quantity/Positive distribution/negative distribution ]

Before describing the environment determination processing executed by the processing unit 70 in the preceding S200, the feature amount, the positive distribution, the negative distribution, the positive probability, and the negative probability used in the environment determination processing will be described.

The feature amount is information extracted from the MF spectrum, the UP spectrum, and the DN spectrum, and information extracted from the MF-FFT spectrum, the UP-FFT spectrum, and the DN-FFT spectrum obtained in the process of generating these spectra. Further, information obtained by combining and calculating these pieces of information may be used. Here, 10 feature quantities will be explained. The 10 feature amounts are referred to as first to tenth feature amounts D1 to D10.

As shown in fig. 5, a region equal to or larger than a predetermined frequency threshold value in the FFT spectrum and the two-dimensional spectrum is referred to as a determination region. The frequency threshold is set according to the vehicle speed, that is, the moving speed of the moving object on which the radar device 10 is mounted. This is because, as the stop has a relative velocity corresponding to the own vehicle velocity, the frequency band to be a determination region in the FMCW spectrum is shifted according to the relative velocity. When the vehicle is moving forward and a stop is detected behind the installation position of the radar device 10, the stop has a positive relative speed. When a stop is detected forward of the installation position of the radar device 10, the stop has a negative relative velocity. The frequency threshold is set to a high frequency to the extent that reflected power from a roadside object or the like located in the vicinity of the host vehicle and on the side does not mix into the determination region. This is because a distant stop located behind the host vehicle is detected in a direction close to the front and back of the host vehicle, but a stop located in the vicinity of and to the side of the host vehicle is detected in the front and lateral direction. The peak detected in the determination region is referred to as a determination peak. The distance corresponding to the frequency threshold corresponds to a threshold distance.

The first feature quantity D1 is the total number of determination peaks on the two-dimensional spectrum.

The second feature quantity D2 is the average power of the determination peak on the two-dimensional spectrum.

The third feature quantity D3 is a power variance of a determination peak on a two-dimensional spectrum.

The fourth feature quantity D4 is the average azimuth of the peak for determination on the two-dimensional spectrum.

The fifth feature quantity D5 is the azimuth variance of the determination peak on the two-dimensional spectrum.

The sixth feature quantity D6 is an average value (hereinafter, average arrival wave number) of the number of determination peaks separated from one determination peak on the two-dimensional spectrum by azimuth expansion for each determination peak on the FFT spectrum.

The seventh feature quantity D7 is the average power of the decision peak on the FFT spectrum.

The eighth feature quantity D8 is a power variance of a decision peak on the FFT spectrum.

The ninth feature quantity D9 is the power ratio of the FFT spectrum to the noise floor (i.e., S/N).

The tenth feature quantity D10 is the variance of the power ratio of the FFT spectrum to the noise floor.

Here, the orientation of the peak for determination used in the fourth feature quantity D4 is the difference with respect to the rear-front direction of the host vehicle. However, instead of the difference with respect to the rear-front direction, the orientation of the determination peak may simply be used as the fourth feature amount D4. The third feature quantity D3, the fifth feature quantity D5, the eighth feature quantity D8, and the tenth feature quantity D10 are not actually the variance itself, but values obtained by multiplying the reciprocal of the variance by a coefficient are used. Note that, although the average power and the power variance of the peak for determination are shown as the feature extracted from the two-dimensional spectrum, the power ratio and the power variance with the noise floor may be used instead of the simple average power and the simple power variance as in the ninth feature D9 and the tenth feature, or the two-dimensional spectrum may be used instead of the peak for determination. When the spectrum is used instead of the peak as a target for calculating the power, the power ratio, and the power variance, the point in the spectrum at which the power other than the peak is used differs. In the first feature quantity D1, the total number of determination peaks on the two-dimensional spectrum is used, but the total number of determination peaks on the FFT spectrum may be used.

When the i-th feature amount Di is acquired, the distribution indicating the probability that the host vehicle travels in the specific environment is defined as a positive distribution, and the distribution indicating the probability that the host vehicle travels in a non-specific environment other than the specific environment is defined as a negative distribution. The positive distribution and the negative distribution are generated based on learning data obtained by manually associating feature amounts collected in advance based on the traveling of the mobile object with the feature amounts acquired in either one of the specific environment and the non-specific environment.

The positive distribution is generated by generating a histogram using a plurality of i-th feature quantities Di collected while traveling in a specific environment, and if the feature quantities are feature quantities conforming to a normal distribution, the positive distribution is further generated by obtaining the normal distribution. Similarly, the negative distribution is generated by generating a histogram using a plurality of i-th feature amounts Di collected when traveling in the unspecific environment, and further by obtaining a normal distribution if the feature amount is a feature amount conforming to the normal distribution. In other words, when calculating the probability using the positive distribution and the negative distribution, the probability is calculated using the normal distribution if the feature quantity is in accordance with the normal distribution, and the probability is calculated using the histogram if the feature quantity is not in accordance with the normal distribution. Fig. 6 to 11 illustrate histograms of the first to tenth feature amounts D1 to D10. In particular, a normal distribution is also exemplified for the first feature quantity D1 and the second feature quantity D2.

The positive probability p (p) is a probability that a specific environment appears during traveling, and the negative probability p (n) is a probability that a non-specific environment appears during traveling.

The positive distribution and the negative distribution, and the positive probability p (p) and the negative probability p (n) generated for each of the first feature quantity D1 through the tenth feature quantity D10 are stored in the memory 72. The area in the memory 72, in which the positive distribution and the negative distribution are stored, corresponds to a distribution storage unit.

[ 1-2-3. environmental judgment treatment ]

The environment determination process executed by the processing unit 70 will be described with reference to the flowchart of fig. 12.

In S410, the processing unit 70 calculates the first to tenth feature amounts D1 to D10 based on the determination peak belonging to the determination region among the object peaks detected in the preceding S170. In the calculation of each feature amount, one predetermined FFT spectrum or two-dimensional spectrum may be used, or a plurality of FFT spectra or two-dimensional spectra may be used. When a plurality of frequency spectrums are used, the feature value may be calculated for each frequency spectrum, or the average of the calculated values in each frequency spectrum may be used as the feature value.

In the next S420, the processing unit 70 calculates a specific environment probability P (Di | P) and a non-specific environment probability P (Di | N) for each of the first to tenth feature amounts D1 to D10. Specifically, the probability corresponding to the value of the i-th feature amount Di calculated in S410 is read from the positive distribution and the negative distribution stored in the memory 72.

In next S430, the processing unit 70 calculates the integration probability PP according to expressions (1) to (3). In this equation, the probability that the environment in which the host vehicle is traveling is a specific environment is calculated using naive bayes that assumes that the specific environment probability P (Di | P) and the unspecific environment probability P (Di | N) are independent phenomena and applies bayesian inference.

[ formula 1]

(2) The expression represents the probability of synthesizing the specific environment probabilities P (Di | P) for all the feature values D1 to D10, and the expression (3) represents the probability of synthesizing the unspecific environment probabilities P (Di | N) for all the feature values D1 to D10. However, instead of expressions (2) and (3), expressions (4) and (5) may be used to reduce the amount of computation and to suppress the loss of numbers in software.

[ formula 2]

In next S440, the processing unit 70 determines whether the integration probability PP calculated in S430 is greater than a threshold TH set in advance. If the integrated probability PP is greater than the threshold TH, the process proceeds to S450, and if the integrated probability PP is less than or equal to the threshold TH, the process proceeds to S460. The threshold TH is determined by an experiment or the like so that the probability of erroneously determining the non-specific environment as the specific environment is reduced as much as possible.

In S450, the processing unit 70 determines that the environment in which the vehicle is traveling is the specific environment, and ends the environment determination process.

In S460, the processing unit 70 determines that the environment in which the vehicle is traveling is a nonspecific environment, and ends the environment determination process.

In the environment determination process, S410 corresponds to a feature amount calculation unit, and S420 to S460 correspond to an environment determination unit.

[ 1-3. Effect ]

According to the first embodiment described above in detail, the following effects are obtained.

(1a) The radar device 10 extracts a plurality of feature quantities D1 to D10 from an FFT spectrum and a two-dimensional spectrum calculated from received signals based on a plurality of modulation schemes. Then, whether or not the vehicle is in the specific environment is determined using the integrated probability PP obtained by integrating the specific environment probability P (Di | P) and the unspecified environment probability P (Di | N) calculated for each of the feature amounts D1 to D10. Therefore, according to the radar device 10, it is possible to determine whether or not the environment is a specific environment in consideration of the plurality of elements expressed by the feature values D1 to D10, and therefore, it is possible to improve the determination accuracy. As a result, the ghost determination can be performed only when the radar apparatus is in a specific environment, and the processing load of the radar apparatus 10 can be reduced.

(1b) The radar device 10 uses naive bayes assuming that each phenomenon is independent when integrating the specific environment probability P (Di | P) and the unspecific environment probability P (Di | N). Therefore, the addition and deletion of the feature quantity Di can be easily performed, and the determination accuracy of whether or not the specific environment is present can be appropriately adjusted.

(1c) In the radar device 10, the threshold TH for the evaluation integration probability PP is set to suppress the non-specific environment from being erroneously determined as the specific environment as much as possible. In other words, if the processing for removing the ghost is performed even in the non-specific environment, the actual target object may be erroneously determined as the ghost and the danger may be ignored, but the occurrence of this can be suppressed. On the other hand, even if the specific environment is erroneously determined as the non-specific environment, the ghost is not removed, but only false alarm based on the ghost increases and the risk is not ignored, so that the threshold TH is set on the safety side in consideration of these. However, the threshold TH may be variably set according to the running environment, the mileage information such as the vehicle speed and the steering angle, and the running environment.

[2. second embodiment ]

[ 2-1 ] different from the first embodiment ]

The second embodiment has the same basic configuration as the first embodiment, and therefore, the following description will discuss different points. Note that the same reference numerals as those in the first embodiment denote the same structures, and the above description is referred to.

In the first embodiment, the determination of the specific environment uses the integrated probability PP. In contrast, the second embodiment differs from the first embodiment in points using the logarithmic bayesian ratio. In other words, the contents of the environment determination processing are partially different.

[ 2-2. treatment ]

The virtual image determination processing of the second embodiment, which is executed by the processing unit 70 instead of the environment determination processing of the first embodiment shown in fig. 12, will be described with reference to the flowchart of fig. 13. The processing in S410 to S420 and S450 to S460 is the same as that in the first embodiment, and therefore, the description thereof is omitted.

In S432 following S420, processing section 70 calculates a logarithmic bayesian ratio LB according to expressions (6) and (7). Here, logP (D | P) and logP (D | N) are calculated by using the above-mentioned expressions (4) and (5).

LB＝TH+log P(D|P)－log P(D|N) (6)

TH＝log P(P)－log P(N)+A (7)

(7) In the formula, a is a constant set by experiment, and may be 0. TH is a threshold value set so that if LB is a positive value, it can be determined as a specific environment, and if LB is a negative value, it can be determined as a non-specific environment.

In the next S434, the processing unit 70 performs a filtering process of suppressing a sharp change in value on the logarithmic bayesian ratio LB. The logarithmic bayesian ratio calculated in S432 is denoted by LB [ n ], the logarithmic bayesian ratio after the filtering process is denoted by LBf [ n ], and the logarithmic bayesian ratio after the filtering process calculated in the previous processing cycle is denoted by LBf [ n-1 ]. In addition, the coefficient α is a real number of 0 < α ≦ 1. In the filtering process, the operation shown in expression (8) is performed. The filtered logarithmic Bayesian ratio LBf [ n ] is also recorded as only logarithmic Bayesian ratio LBf.

LBf[n]＝α×LB[n]+(1－α)LBf[n－1] (8)

In next S442, the processing unit 70 determines whether the logarithmic bayesian ratio LBf is greater than 0. If the logarithmic bayesian ratio LBf is greater than 0, the environment in which the vehicle is traveling is a specific environment, and the process proceeds to S450, and if the logarithmic bayesian ratio LBf is not greater than 0, the environment in which the vehicle is traveling is a nonspecific environment, and the process proceeds to S460.

[ 2-3. operation example ]

Fig. 14 and 15 each show the upper graph showing the phantom probability PG calculated by expression (4), and the lower graph showing the result of plotting the logarithmic bayesian ratio LB calculated by expression (6) before the filtering process and the logarithmic bayesian ratio LBf calculated by expression (8) for each processing cycle. Fig. 14 shows a case where the environment in which the host vehicle travels is a non-specific environment, and fig. 15 shows a case where the environment in which the host vehicle travels is a specific environment.

As shown in fig. 14, in the 25 th processing cycle, due to the interference, the integrated probability PP becomes a large value exceeding 50% even in the non-specific environment, and the logarithmic bayesian ratio LB becomes a positive value. In other words, when the integrated probability PP and the logarithmic bayesian ratio LB before the filtering process are used to determine whether or not the environment is a specific environment, there is a possibility that an erroneous determination result is obtained. On the other hand, the logarithmic bayesian ratio LBf after the filtering process is still negative, and a correct determination result is obtained.

Further, the threshold TH can be changed by adjusting the parameter a of the formula (7). For example, when a is set to 0, the probability of erroneously determining the specific environment as the non-specific environment is equal to the probability of inversely erroneously determining the non-specific environment as the specific environment.

For example, consider a case where the in-vehicle system 1 operates as a system that gives an alarm to a target object approaching the own vehicle. In this case, there is a higher possibility that the system is not operated and becomes a problem when the specific environment is erroneously determined as the non-specific environment (e.g., a general road) than when the system is unnecessarily operated when the non-specific environment is erroneously determined as the specific environment (e.g., a tunnel). In other words, it is more problematic to not issue an alarm in a dangerous situation than to issue an alarm in a non-dangerous situation.

For example, the in-vehicle system 1 may be operated as a system that detects and deletes a ghost (that is, a virtual image) from the target object candidates and that warns a target object close to the host vehicle. In this case, if the system is unnecessarily operated by erroneously determining the unspecific environment as the specific environment, there is a possibility that the target object to be detected is unnecessarily deleted.

Therefore, in order to suppress the occurrence of these cases, the probability of erroneously determining the unspecified environment as the specified environment can be further reduced by setting a > 0, that is, shifting the threshold TH to the positive side in fig. 14. In addition, the probability of erroneously determining the specific environment as the non-specific environment may be further reduced by the processing realized by the in-vehicle system 1, and conversely, by setting a < 0 and shifting the threshold TH to the negative side.

[ 2-4. Effect ]

According to the second embodiment described in detail above, the effects (1a) to (1c) of the first embodiment described above are obtained, and the following effects are obtained.

(2a) The logarithmic Bayesian ratio LBf is calculated by performing a filtering process in which the logarithmic Bayesian ratio LB [ n ] obtained in the current processing period before the filtering process and the logarithmic Bayesian ratio LBf [ n-1 ] obtained in the previous processing period after the filtering process are mixed by using a coefficient alpha. Therefore, even if the abnormal logarithmic bayesian ratio LB [ n ] is suddenly calculated due to the disturbance, the determination as to whether or not the specific environment is present is not immediately followed, and a stable determination result can be obtained.

(2b) By using the logarithmic bayesian ratio LBf, the amount of computation can be further reduced as compared with the case of using the integrated probability PP.

(2c) The threshold TH is provided in the calculation of the logarithmic bayesian ratio LB, and the threshold TH can be easily adjusted.

[3 ] third embodiment ]

[ 3-1 ] different from the first embodiment ]

The third embodiment has the same basic configuration as the first embodiment, and therefore, the following description will discuss different points. Note that the same reference numerals as those in the first embodiment denote the same structures, and the above description is referred to.

In the first embodiment, when determining whether or not the environment is a specific environment, the positive distribution and the negative distribution stored in advance in the memory 72 are used, and the values thereof are not changed. In contrast, the third embodiment differs from the first embodiment in that the positive distribution and the negative distribution stored in the memory 72 are appropriately updated by the learning process. In other words, the content of the target object detection processing is partially different.

In the present embodiment, as the operation mode, there is a learning mode in which a learning process is executed. Even before the learning process is executed, initial values of the positive distribution and the negative distribution are stored in the memory 72.

[ 3-2. target detection treatment ]

The target object detection processing of the third embodiment, which is executed by the processing unit 70 instead of the target object detection processing of the first embodiment shown in fig. 3, will be described with reference to the flowchart of fig. 16. Note that the target object detection processing of the present embodiment is the same as that of the first embodiment except that S175 and S240 are added, and therefore, the difference will be described.

In S175 following S170, the processing unit 70 determines whether or not the operation mode of the radar device 10 is the learning mode, and if the operation mode is the learning mode, the processing proceeds to S240, and if the operation mode is not the learning mode, the processing proceeds to S180. The operation mode is set by an input operation from an input device provided outside the processing unit 70.

In S240, the processing unit 70 executes the learning process and ends the process. The learning process is a process of generating and updating a probability distribution for determining whether or not the environment is a specific environment.

[ 3-3. learning Process ]

The learning process executed by the processing unit 70 in S250 will be described using the flowchart of fig. 17.

In S310, the processing unit 70 calculates the first to tenth feature amounts D1 to D10 based on the determination peak belonging to the determination region among the object peaks detected in the preceding S170.

In the calculation of each feature amount, one predetermined FFT spectrum or two-dimensional spectrum may be used, or a plurality of FFT spectra or two-dimensional spectra may be used. When a plurality of frequency spectrums are used, the feature value may be calculated for each frequency spectrum, or the average of the calculated values in each frequency spectrum may be used as the feature value.

In S320, the processing unit 70 stores the feature amount calculated in S310 in the memory 72 in association with the teacher data. The teacher data is data indicating whether or not the environment in which the host vehicle is traveling is a specific environment, and is set by sequentially inputting, to the processing unit 70, the result of the determination of whether the occupant of the vehicle is the specific environment or the unspecific environment, for example, in the learning mode. In addition, when the system includes a device capable of determining in what environment the vehicle is traveling by using a navigation system, a camera, a satellite positioning system (for example, GPS), or the like (for example, in the case of a vehicle dedicated to data collection), teacher data may be acquired from these devices. Further, as shown in the histograms in fig. 6 to 11, the feature quantities (hereinafter, feature quantities with teacher data) having a correspondence relationship with the teacher data are accumulated for each feature quantity and for each teacher data (i.e., for each specific environment).

In next S330, the processing unit 70 determines whether or not the timing is the update timing, and if the timing is the update timing, the processing proceeds to S340, and if not, the processing is terminated. The update timing may be a timing that ensures that the feature quantity with teacher data accumulated in the memory 72 by the processing of S320 reaches the number necessary for generation and update of the positive distribution and the negative distribution.

In S340, the processing unit 70 generates a positive distribution and a negative distribution for each feature amount type based on the accumulated feature amounts with teacher data, and generates a positive probability p (p) and a negative probability p (n). Specifically, the positive distribution and the negative distribution shown in fig. 6 and 7 are generated by expressing the distribution of the feature amounts shown in the histograms in fig. 6 to 11 by a normal distribution. Although not shown in the drawings, the histograms of fig. 8 to 11 similarly generate positive and negative distributions. The positive probability p (p) and the negative probability p (n) are calculated using results obtained by summarizing the number of feature quantities collected in a specific environment and the number of feature quantities collected in a non-specific environment. However, the positive probability p (p) and the negative probability p (n) may be fixed values.

In the next S350, the processing unit 70 updates the positive distribution and the negative distribution stored in the memory 72 in accordance with the positive distribution and the negative distribution generated in S340, and ends the processing.

As shown in fig. 6 to 11, the first feature quantity D1, the fourth feature quantity D4 to the sixth feature quantity D6, the eighth feature quantity D8, and the tenth feature quantity D10 tend to have values smaller in the specific environment than in the non-specific environment. In addition, the second feature amount D2, the seventh feature amount D7, and the ninth feature amount D9 tend to have a larger value in the specific environment than in the non-specific environment.

As shown in fig. 8, as the fourth feature amount D4, an average of the azimuth differences of the peaks with respect to the rear-front direction is used, but the average azimuth of the peaks may be used instead of the fourth feature amount D4 or in addition to the first to eighth feature amounts D1 to D8.

In the learning process, S310 corresponds to a feature amount calculation unit, S320 corresponds to a collection unit, and S330 to S350 correspond to learning units.

[ 3-4. Effect ]

According to the third embodiment described in detail above, the effects (1a) to (1c) of the first embodiment described above are obtained, and the following effects are obtained.

(3a) By using the learning mode, the positive distribution and the negative distribution stored in the memory 72 can be updated, and therefore the accuracy of determining whether or not the environment is a specific environment can be sequentially improved.

[4 ] other embodiments ]

While the embodiments of the present disclosure have been described above, the present disclosure is not limited to the embodiments described above, and can be implemented in various modifications.

(4a) In the above embodiment, the environment determination process is performed using the first to tenth feature amounts D1 to D10, but the present disclosure is not limited thereto. For example, the information on the surroundings of the host vehicle obtained from the image of the onboard camera and the information obtained using the position information of the host vehicle and the map information may be combined to determine whether or not the environment is a specific environment.

(4b) In the above embodiment, the first to tenth feature amounts D1 to D10 are used as the feature amounts used for calculating the log bayesian ratio LB, but any feature amount may be used as long as a clear difference is generated between positive and negative distributions. The number of feature quantities used for calculating the integration probability PP is not limited to 10, and may be 1 to 9, or 11 or more.

(4c) In the above embodiment, when the log probabilities, i.e., log P (D | P) and log P (D | N), are obtained by using the expressions (4) and (5), it is easy to obtain very small values. Wherein the final calculated integrated virtual image probability PG is obtained by the ratio of these values. Therefore, the maximum value of logP (Di | R) and logP (D | I) may be MAX, and the result of subtracting the maximum value from each may be used as the log probability.

(4d) In the above-described embodiment, the radar device 10 is disposed so as to include the rear side along the advancing direction of the vehicle within the detection range, but the present disclosure is not limited thereto. For example, the radar device 10 may be arranged so as to include the front in the traveling direction of the vehicle within the detection range. In this case, for example, the azimuth used for the feature amount may be calculated using the front frontal direction as the azimuth difference from the reference direction.

(4d) The processing unit 70 and its method described in the present disclosure may also be implemented by a special purpose computer provided by constituting a processor and a memory programmed to execute one or more functions embodied by a computer program. Alternatively, the processing unit 70 and the method thereof described in the present disclosure may be implemented by a special purpose computer provided by a processor configured by one or more dedicated hardware logic circuits. Alternatively, the processing unit 70 and the method thereof described in the present disclosure may be implemented by one or more special purpose computers configured by a combination of a processor programmed to execute one or more functions, a memory, and a processor configured by one or more hardware logic circuits. The computer program may be stored in a non-transitory tangible recording medium that can be read by a computer as instructions to be executed by the computer. The method for implementing the functions of each part included in the processing unit 70 does not necessarily include software, and all the functions may be implemented by using one or more hardware.

(4e) The plurality of components may realize a plurality of functions of one component in the above embodiments, or a plurality of components may realize one function of one component. Further, a plurality of functions provided by a plurality of components may be realized by one component, or one function realized by a plurality of components may be realized by one component. In addition, a part of the structure of the above embodiment may be omitted. In addition, at least a part of the structure of the above embodiment may be added to or replaced with the structure of the other above embodiment.

(4f) The present disclosure can be realized in various forms other than the radar device 10 described above, such as a system having the radar device 10 as a component, a program for causing a computer to function as the radar device 10, a non-transitory tangible recording medium such as a semiconductor memory in which the program is recorded, an environment determination method, and the like.