JP2014219249A - Sensor device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a sensor device capable of reducing erroneous detection caused from a movement of an object other than a detection target.SOLUTION: A signal processor 2 of a sensor device Se includes: a frequency analysis unit 5 that converts a sensor signal output from an A/D converter 4 into a signal in a frequency region to extract as signals for each filter bank in a filter bank group each having different frequency band; a standardization unit 6 that standardizes the field strength of the signals each has passed through the filter banks; a first recognition unit 7a that performs recognition processing to recognize an object based on the frequency distribution determined based on a standardization field strength of each filter bank; a background signal removal unit 10 that removes a background signal from the signals passed through the filter bank; and a second recognition unit 7b that, when the first recognition unit 7a identifies the object, identifies a steady background signal based on at least one of a frequency distribution characteristic which is determined from the standardization field strength of each filter bank and a characteristic of time-continuous frequency distribution.

Description

  The present invention relates to a sensor device including a radio wave sensor and a signal processing device that processes a sensor signal from the radio wave sensor.

  Conventionally, an illumination system configured as shown in FIG. 27 has been proposed (Patent Document 1). This lighting system includes an object detection device 101 including a sensor 110 that detects the presence or absence of a detection target in a detection area and outputs a sensor signal, and a lighting fixture 102 whose lighting state is controlled by the object detection device 101. I have.

  The sensor 110 transmits the millimeter wave toward the detection area, receives the millimeter wave reflected by the detection object moving in the detection area, and determines the frequency difference between the transmitted millimeter wave and the received millimeter wave. This is a millimeter wave sensor that outputs a sensor signal having a corresponding Doppler frequency.

  The object detection apparatus 101 divides a sensor signal output from the sensor 110 into a plurality of frequency bands and amplifies the signal for each frequency band, and compares the output of the amplification circuit 111 with a predetermined threshold value to detect the detection target object. And a determination unit 112 that determines presence or absence. Further, the object detection apparatus 101 includes an illumination control unit 113 that controls the lighting state of the lighting fixture 102 according to the determination result of the determination unit 112.

  Further, the object detection apparatus 101 includes a frequency analysis unit 114 that detects the intensity of each sensor signal output from the sensor 110 for each frequency. In addition, the object detection apparatus 101 includes a noise removal unit (noise determination unit 115 and switching circuit 116) that reduces the influence of noise of a specific frequency that is constantly generated using the analysis result of the frequency analysis unit 114. Here, as the frequency analysis unit 114, an FFT (Fast Fourier Transform) analyzer is used. The determination unit 112, the illumination control unit 113, and the noise removal unit are included in a control block 117 whose main configuration is a microcomputer. The amplifier circuit 111 constitutes a signal processing unit that outputs sensor signals for each predetermined frequency band. Note that Patent Document 1 describes that the signal processing unit may be configured using an FFT analyzer, a digital filter, or the like.

  The amplifier circuit 111 includes a plurality of amplifiers 118 using operational amplifiers. By adjusting various parameters of the circuits constituting each amplifier 118, it is possible to set a frequency band for amplifying a signal by each amplifier 118. It has become. That is, each amplifier 118 also functions as a band-pass filter that passes a signal in a specific frequency band. Thus, in the amplifier circuit 111, the sensor signals are divided into a plurality of frequency bands by a plurality of amplifiers 118 connected in parallel, and the signals in the respective frequency bands are respectively amplified by the amplifiers 118 and output individually.

The determination unit 112 A / D converts the output of the amplifier 118 into a digital value, and has a comparator 119 for comparing with a predetermined threshold value for each amplifier 118, and determines whether or not a detection target is present. In the comparator 119, the threshold value is individually set for each pass band (that is, for each amplifier 118), and outputs an H level signal when the output of the amplifier 118 is outside the range defined by the threshold value. Here, the threshold value Vth of each pass band set in the initial state (shipment state) is the output value V of each amplifier 118 measured within a certain time in a state where there is no reflection of electromagnetic waves as in an anechoic chamber or the like. Using the maximum value Vpp _{ini of the} peak-to-peak Vpp and the average value Vavg of the output value V, the value is expressed as Vth = Vavg ± Vpp _ini . The determination unit 112 includes a logical sum circuit 120 that performs a logical sum of the comparison results. If there is at least one H level signal, the determination unit 112 outputs a detection signal indicating a “detection state” in which a detection target exists. On the other hand, a detection signal indicating a “non-detection state” in which there is no detection target is output from the OR circuit 120 if all are at the L level. The detection signal is “1” in the detection state and “0” in the non-detection state.

  The noise removing unit determines from the output of the frequency analyzing unit 114 whether or not there is a noise having a specific frequency that is constantly generated, and each amplifier 118 for the determining unit 112 according to the determination result of the noise determining unit 115. And a switching circuit 116 for switching the output state.

  The switching circuit 116 includes switches 121 inserted between the amplifiers 118 of the amplifier circuit 111 and the comparators 119 of the determination unit 112. In an initial state, all the switches 121 are turned on. Then, each switch 121 is individually turned on / off by the output from the noise determination unit 115, so that the output of each amplifier 118 to the determination unit 121 is turned on / off individually. That is, in the switching circuit 116, the output of the amplifier 118 can be invalidated by turning off the switch 121 corresponding to the amplifier 118 of an arbitrary pass band by the output from the noise determination unit 115.

  In the noise determination unit 115, the signal strength (voltage strength) of the sensor signal for each frequency (frequency component) output from the frequency analysis unit 114 is read and stored in a memory (not shown), and the stored data is used for steady state. The presence or absence of specific frequency noise is determined.

  When the noise determination unit 115 determines that noise of a specific frequency is constantly generated, the switch 121 between the amplifier 118 having the pass band including the noise and the determination unit 112 is turned off. The switching circuit 116 is controlled. Thereby, when the noise of a specific frequency has generate | occur | produced regularly, the output of the amplifier circuit 111 with respect to the determination part 112 becomes invalid about the frequency band containing the said noise. Here, the ON / OFF state of the switch 121 is updated every time the noise determination unit 115 determines “normal”.

JP 2011-47779 A

  In the object detection device 101, it is considered that a portion excluding the sensor 110 and the illumination control unit 113 constitutes a signal processing device that performs signal processing on the sensor signal of the sensor 110 that is a millimeter wave sensor. However, in the object detection device 101, for example, when used outdoors, it is caused by the movement of an object other than the detection target (detection target) (for example, the movement of a tree branch or leaf, the movement of an electric wire, etc.). There is a possibility that an object other than the detection target is erroneously detected as the detection target object.

  The present invention has been made in view of the above-described reasons, and an object thereof is to provide a sensor device capable of reducing erroneous detection caused by the movement of an object other than a detection target.

  The sensor device of the present invention includes a radio wave sensor that transmits radio waves and receives radio waves reflected by an object, and a signal processing device that performs signal processing on sensor signals corresponding to the movement of the object output from the radio wave sensor. The signal processing device includes: an amplifying unit that amplifies a sensor signal output from the radio wave sensor; an A / D converting unit that converts the sensor signal amplified by the amplifying unit into a digital sensor signal; A frequency analysis means for converting the sensor signal output from the A / D conversion section into a signal in the frequency domain and extracting the signal as a signal for each filter bank in a group of filter banks having different frequency bands; and the frequency analysis means The total sum of the received signals or the sum of the intensities of the signals that have passed through the predetermined plurality of filter banks passed through each of the filter banks. Normalization means for normalizing the intensity of the signal and outputting it as normalized intensity, and at least one of a frequency distribution determined from the normalized intensity for each filter bank output from the normalization means or a component ratio of the normalized intensity First recognition means for performing recognition processing for identifying the object, background signal removal means for removing a background signal from the signal that has passed through the filter bank, and operation when the object is identified by the first recognition means. The stationary background signal is identified by at least one of the characteristics of the frequency distribution determined from the normalized strength for each filter bank output from the normalizing means and the temporal continuity characteristics of the frequency distribution. A second recognizing means.

  In this sensor device, it is preferable that the second recognition unit periodically recognizes a stationary background signal and updates the background signal.

  In this sensor device, the frequency analysis unit has a function of converting the sensor signal output from the A / D conversion unit into a signal in the frequency domain by performing discrete cosine transform, and each of the filter banks A plurality of frequency bins, and the signal processing device includes a smoothing processing unit between the frequency analyzing unit and the normalizing unit, and the smoothing processing unit includes the frequency for each filter bank. It has at least one of a function for smoothing the signal strength for each bin in the frequency domain and a function for smoothing the signal strength for each frequency bin in the time axis direction for each filter bank. preferable.

  In this sensor device, it is preferable that the background signal removing unit removes the background signal by subtracting the background signal from a signal that has passed through the filter bank.

  In this sensor device, it is preferable that the background signal removing unit removes, as the background signal, an average value on the time axis of a plurality of signals for each filter bank obtained in advance.

  In this sensor device, each of the filter banks has a plurality of frequency bins, the signal processing device sets the frequency bin in which the background signal is steadily included as a specific frequency bin, and the background signal removing means It is preferable that the background signal is removed by invalidating the signal of the specific frequency bin and complementing with the signal intensity estimated from the intensity of the signal of the frequency bin adjacent to the specific frequency bin.

  In this sensor device, it is preferable that the first recognition unit identifies the object by performing pattern recognition by principal component analysis or KL conversion as the recognition processing.

  In this sensor device, it is preferable that the first recognition unit identifies the object by performing recognition processing by multiple regression analysis as the recognition processing.

  In this sensor device, it is preferable that the first recognition unit identifies the object by performing recognition processing using a neural network as the recognition processing.

  In this sensor device, it is preferable that the first recognition unit determines an identification result by majority decision based on an odd number of recognition processing results on a time axis.

  In this sensor device, the signal processing device performs the recognition processing by the first recognition device only when the sum of the intensities of the signals before normalization by the normalization device is greater than or equal to a predetermined value. It is preferable to validate the result of the recognition processing by one recognition means.

  In the sensor device of the present invention, it is possible to reduce false detection caused by the movement of an object other than the detection target.

FIG. 1 is a block diagram of a sensor device according to an embodiment. FIG. 2 is an explanatory diagram of normalization means in the sensor device of the embodiment. FIG. 3 is an explanatory diagram of smoothing processing means in the sensor device of the embodiment. FIG. 4 is an explanatory diagram of an example of a background signal removing unit in the sensor device of the embodiment. FIG. 5 is an explanatory diagram of another example of the background signal removing unit in the sensor device of the embodiment. FIG. 6 is an explanatory diagram of still another example of the background signal removing unit in the sensor device of the embodiment. FIG. 7 is an explanatory diagram of recognition processing by principal component analysis of the signal processing device in the sensor device of the embodiment. FIG. 8 is an explanatory diagram of a usage pattern of the sensor device of the embodiment. FIG. 9 is a waveform diagram of a sensor signal from a radio wave sensor in the sensor device of the embodiment. FIG. 10 is an explanatory diagram of the output of the normalization means in the sensor device of the embodiment. FIG. 11 is an explanatory diagram of an output signal of the sensor device of the embodiment. FIG. 12 is an explanatory diagram of a usage pattern of the sensor device of the embodiment. FIG. 13 is a waveform diagram of a sensor signal from a radio wave sensor in the sensor device of the embodiment. FIG. 14 is an explanatory diagram of an output signal of the sensor device of the embodiment. FIG. 15 is an explanatory diagram of the output of the normalization means in the sensor device of the embodiment. FIG. 16 is an explanatory diagram of an output signal of the sensor device according to the embodiment. FIG. 17 is an explanatory diagram of identification processing by multiple regression analysis of the signal processing device in the sensor device of the embodiment. FIG. 18 is another explanatory diagram of identification processing by multiple regression analysis of the signal processing device in the sensor device of the embodiment. FIG. 19 is an explanatory diagram of the majority decision by the first recognition unit of the signal processing device in the sensor device of the embodiment. FIG. 20 is an explanatory diagram of a signal processing device in the sensor device of the embodiment. FIG. 21 is a schematic configuration diagram of a neural network of the signal processing device in the sensor device of the embodiment. FIG. 22 is a flowchart for explaining the operation of the signal processing device in the sensor device of the embodiment. FIG. 23 is an explanatory diagram of the output of the normalization means in the sensor device of the embodiment. FIG. 24 is an explanatory diagram of the output of the normalization means in the sensor device of the embodiment. FIG. 25 is an explanatory diagram of the output of the normalization means in the sensor device of the embodiment. FIG. 26 is an explanatory diagram of the output of the normalization means in the sensor device of the embodiment. FIG. 27 is a block diagram showing a configuration of a conventional lighting system.

  Below, sensor device Se of this embodiment is explained based on Drawing 1-Drawing 26.

  The sensor device Se includes a radio wave sensor 1 that transmits a radio wave and receives a radio wave reflected by an object, and a signal processing device 2 that performs signal processing on a sensor signal output from the radio wave sensor 1.

  The signal processing device 2 includes an amplification unit 3 that amplifies the sensor signal output from the radio wave sensor 1, and an A / D conversion unit 4 that converts the sensor signal amplified by the amplification unit 3 into a digital sensor signal and outputs the digital sensor signal. It is equipped with. Further, the signal processing device 2 converts the sensor signal output from the A / D conversion unit 4 into a frequency domain signal (frequency axis signal), and a filter bank 5a (FIG. 2A) having a different frequency band. Frequency analysis means 5 for extracting as a signal for each filter bank 5a in the reference group. Further, the signal processing device 2 uses the sum of the signals extracted by the frequency analysis means 5 or the sum of the intensities of the signals that have passed through a predetermined plurality of (for example, four on the low frequency side) filter banks 5a. Normalization means 6 is provided for normalizing the intensity of the signal that has passed through each of the 5a and outputting the normalized intensity. Further, the signal processing device 2 includes first recognition means 7 a that performs recognition processing for recognizing an object based on a frequency distribution determined from the normalized intensity for each filter bank 5 a output from the normalization means 6. The signal processing device 2 includes a background signal removing unit 10 that removes a background signal from a signal that has passed through the filter bank 5a, and a second recognition unit 7b that operates when the object is identified by the first recognition unit 7a. . The second recognizing means 7b is stationary by at least one of the characteristics of the frequency distribution determined from the normalized strength for each filter bank 5a output from the normalizing means 6 and the characteristics of temporal continuity of the frequency distribution. Identify the background signal. Therefore, the sensor device Se can reduce false detection caused by the movement of an object other than the detection target.

  Below, each component of sensor device Se is explained in detail.

  For example, a Doppler sensor is preferably used as the radio wave sensor 1. For example, the Doppler sensor transmits a radio wave of a predetermined frequency toward the detection area, receives a radio wave reflected by an object moving in the detection area, and sets the difference between the frequency of the transmitted radio wave and the received radio wave. A sensor signal having a corresponding Doppler frequency is output. Therefore, the sensor signal output from the radio wave sensor 1 is an analog time axis signal corresponding to the movement of the object.

  A Doppler sensor is a frequency that is equivalent to the difference between the frequency of the transmitter that transmits the radio wave toward the detection area, the receiver that receives the radio wave reflected by the object in the detection area, and the received radio wave. And a mixer for outputting the sensor signal. The transmitter includes a transmission antenna. The receiver includes a receiving antenna. The radio wave transmitted from the transmitter can be, for example, a millimeter wave having a predetermined frequency of 24.15 GHz. The radio wave transmitted from the transmitter is not limited to a millimeter wave but may be a microwave. Further, the value of the predetermined frequency of the radio wave to be transmitted is not particularly limited. When the object reflecting the radio wave is moving within the detection area, the frequency of the reflected wave is shifted by the Doppler effect. The configuration or the like of the radio wave sensor 1 is not particularly limited, and a two-frequency (multi-frequency) type Doppler sensor, an FMCW (Frequency Modulated Continuous Wave) type Doppler sensor, or the like can also be used.

  The amplifying unit 3 can be configured by an amplifier using an operational amplifier, for example. The configuration of the amplifying unit 3 is not particularly limited.

The A / D conversion unit 4 converts the analog sensor signal amplified by the amplification unit 3 into a digital sensor signal and outputs the digital sensor signal. The A / D converter 4 has a sampling rate set to 1 × 10 ³ sps (sample per second), but the sampling rate is not particularly limited.

  The frequency analysis means 5 sets a predetermined number (for example, 16) of filter banks 5a as a group of filter banks 5a, but the number of filter banks 5a is not particularly limited.

  The frequency analysis means 5 has a function of converting the sensor signal output from the A / D converter 4 into a signal in the frequency domain by performing discrete cosine transform (DCT). Further, as shown in FIG. 2A, each of the filter banks 5a has a plurality (five in the illustrated example) of frequency bins 5b. The frequency bin 5b of the filter bank 5a using DCT is also called a DCT bin. The resolution of each filter bank 5a is determined by the width of the frequency bin 5b (Δf in FIG. 2A). The number of frequency bins 5b in each filter bank 5a is not particularly limited, and may be a plurality other than five or one. The orthogonal transform that converts the sensor signal output from the A / D conversion unit 4 into a frequency domain signal is not limited to DCT, and may be, for example, Fast Fourier Transformation (FFT). The frequency bin 5b of the filter bank 5a using FFT is also referred to as FFT bin. Further, the orthogonal transform that converts the sensor signal output from the A / D converter 4 into a frequency domain signal may be a wavelet transform (WT).

  When each of the filter banks 5a has a plurality of frequency bins 5b, the signal processing device 2 includes a smoothing processing unit 8 between the frequency analysis unit 5 and the normalization unit 6. preferable. The smoothing processing means 8 is also referred to as a function (hereinafter referred to as a “first smoothing processing function”) for smoothing the signal strength of each frequency bin 5b in the frequency domain (frequency axis direction) for each filter bank 5a. ) And a function of smoothing the signal strength of each frequency bin 5b in the time axis direction for each filter bank 5a (hereinafter, also referred to as “second smoothing function”). It is preferable. Thereby, the signal processing device 2 can reduce the influence of noise, and if both are provided, the influence of noise can be further reduced.

The first smoothing processing function of the smoothing processing means 8 can be realized by, for example, an average value filter, a load average filter, a median filter, a load median filter, or the like. When the first smoothing processing function is realized by an average value filter, the first filter bank is counted in order from the lowest frequency as shown in FIGS. 2A and 3A at time t ₁ . Assume that the signal strengths in the five frequency bins 5b of 5a are s1, s2, s3, s4, and s5, respectively. Here, regarding the first filter bank 5a, if the intensity of the signal smoothed by the first smoothing function is m ₁₁ (see FIGS. 2B and 3B), the intensity m ₁₁ Is obtained by the calculation of the following equation (1).
m ₁₁ = (s1 + s2 + s3 + s4 + s5) / 5 (1) Similarly, the signals of the second filter bank 5a, the third filter bank 5a, the fourth filter bank 5a and the fifth filter bank 5a are shown in FIG. ) And m ₂₁ , m ₃₁ , m _41, and m ₅₁ , respectively, as shown in FIG. In short, in the above description, for convenience of explanation, the signal of the jth filter bank 5a (j is a natural number) at time t _i (i is a natural number) on the time axis is smoothed by the first smoothing function. The intensity of the processed signal is represented as m _ji .

The normalizing means 6 normalizes the intensity values of the signals that have passed through the respective filter banks 5a by the sum of the intensities of the signals that have passed through a plurality of predetermined filter banks 5a used for recognition processing in the first recognizing means 7a. . Here, for example, the total number of the filter banks 5a in the frequency analysis means 5 is 16, and the predetermined plurality of filter banks 5a used for the recognition process are counted as the first to fifth five from the lowest frequency. It will be described as being only. Assuming that the normalized strength of the signal strength m ₁₁ passing through the first filter bank 5a at time t ₁ is n ₁₁ (see FIG. 2C), the normalized strength n ₁₁ is It is calculated | required by the calculation of (2) Formula.
n ₁₁ = m ₁₁ / (m ₁₁ + m ₂₁ + m ₃₁ + m ₄₁ + m ₅₁ ) (2) Equation (2) When each filter bank 5a is composed of one frequency bin 5b, the normalizing means 6 uses each filter bank 5a. The intensity of the signal that has passed through each is extracted, and the intensity of the signal that has passed through each filter bank 5a is normalized by the sum of these intensities.

Further, the second smoothing processing function of the smoothing processing means 8 can be realized by, for example, an average value filter, a load average filter, a median filter, a load median filter, or the like. When the second smoothing function is realized by an average value filter for obtaining an average value at a plurality of points (for example, three points) in the time axis direction, as shown in FIG. 3C, the first filter bank 5a If the intensity of the signal smoothed by the second smoothing function is m ₁ , the intensity m ₁ can be obtained by the calculation of the following equation (3).
m ₁ = (m ₁₀ + m ₁₁ + m ₁₂ ) / 3 (3) Similarly, the signals of the second filter bank 5a, the third filter bank 5a, the fourth filter bank 5a, and the fifth filter bank 5a are obtained. M ₂ , m ₃ , m ₄ and m ₅ , m ₂ , m ₃ , m ₄ and m ₅ represent the following formulas (4), (5), (6) and (7, respectively. ) It is calculated | required by each calculation.
m ₂ = (m ₂₀ + m ₂₁ + m ₂₂ ) / 3 (4) Formula m ₃ = (m ₃₀ + m ₃₁ + m ₃₂ ) / 3 (5) Formula m ₄ = (m ₄₀ + m ₄₁ + m ₄₂ ) / 3 (6) Expression m ₅ = (m ₅₀ + m ₅₁ + m ₅₂ ) / 3 (7) In short, in the above description, for the sake of convenience of explanation, the first smoothing is performed on the signal of the n-th filter bank 5a (n is a natural number). The intensity of the signal smoothed by the processing function and further smoothed by the second smoothing function is denoted by _mn .

  Moreover, it is preferable that the signal processing apparatus 2 includes a background signal estimation unit 9 that estimates a background signal (that is, noise) included in a signal output from each filter bank 5a.

  For example, the signal processing device 2 identifies, as operation modes, a first mode in which a background signal is estimated, a second mode in which recognition processing is performed by the first recognition unit 7a, and a stationary background signal by the second recognition unit 7b. And a third mode. Then, the signal processing device 2 switches between the first mode and the second mode every predetermined time (for example, 30 seconds) timed by a timer (not shown) provided in the signal processing device 2. It is preferable to operate in the third mode when an object is identified by the first recognition means 7a in the second mode. Here, the signal processing apparatus 2 operates the background signal estimation means 9 during the first mode period, and after the background signal removal means 10 removes the background signal during the second mode period, the first recognition means 7a. It is preferable to perform a recognition process. The time in the first mode and the time in the second mode are not limited to the same time (for example, 30 seconds), and may be different from each other.

For example, the background signal removing unit 10 may remove the background signal by subtracting the background signal from the signal output from the filter bank 5a. In this case, the background signal removal means 10 is estimated by the background signal estimation means 9 from the intensities of the signals m ₁ , m ₂ ,... (See FIG. 4B) that have passed through the filter banks 5a, for example. Further, it can be constituted by a subtracter for subtracting the intensity b ₁ , b ₂ ,... (See FIG. 4A) of the background signal. FIG. 4C shows the intensity of the signal obtained by subtracting the background signal from the signal between the same filter banks 5a. Here, if the intensity of the signal of the first filter bank 5a from the left and L _1, the intensity L ₁ is obtained by the following equation (8).
L ₁ = m ₁ −b ₁ (8) Similarly, after subtracting the background signal for the second filter bank 5a, the third filter bank 5a, the fourth filter bank 5a, and the fifth filter bank 5a the intensity of the signal, _{respectively,} if L _2, L 3, _{L 4} and _{L 5,} L _2, L 3, _{L 4} and _{L 5,} the following equation (9), (10), (11) It calculates | requires by each calculation of Formula and (12) Formula.
_{L 2 = m 2 -b 2 (} 9) formula _{L 3 = m 3 -b 3 (} 10) formula _{L 4 = m 4 -b 4 (} 11) formula _{L 5 = m 5 -b 5 (} 12) Equation background signal The estimation means 9 may estimate the signal strength obtained for each filter bank 5a as the background signal strength for each filter bank 5a and update it as needed during the first mode. The background signal estimation means 9 may estimate the average value of the intensity of the plurality of signals obtained for each filter bank 5a as the intensity of the background signal for each filter bank 5a in the first mode. Good. That is, the background signal estimation means 9 may use an average value on the time axis of a plurality of signals for each filter bank 5a obtained in advance as the background signal. Thereby, the background signal estimation means 9 can improve the estimation accuracy of the background signal.

Further, the background signal removing means 10 may use the signal immediately before each filter bank 5a as the background signal in the first mode. Here, the signal processing device 2 may have a function of removing the background signal by subtracting the immediately preceding signal on the time axis before each signal is normalized by the normalizing means 6. In short, the background signal removing unit 10 has a function of removing the background signal by subtracting the signal one sample before on the time axis from the signal to be normalized for the signal that has passed through each filter bank 5a. You may make it have. In this case, for example, as shown in FIG. 5, the signals of the respective filter banks 5a at the time t ₁ to be subjected to normalization processing are expressed as m ₁ (t ₁ ), m ₂ (t ₁ ), m ₃ (t ₁ ), m ₄ (t ₁ ), and m ₅ (t ₁ ), and the signals at time t ₀ immediately before are m ₁ (t ₀ ), m ₂ (t ₀ ), m ₃ (t ₀ ), m ₄ (t ₀ ) and m ₅ (t ₀ ), and L ₁ , L ₂ , L ₃ , L ₄ and L ₅ are the signal strengths after subtraction, L ₁ , L ₂ , L ₃ , L ₄ and _{L 5,} (13) below, (14), (15) is obtained by (16) and (17) each operation.
L ₁ = m ₁ (t ₁ ) −m ₁ (t ₀ ) (13) Formula L ₂ = m ₂ (t ₁ ) −m ₂ (t ₀ ) (14) Formula L ₁ = m ₃ (t ₁ ) − m ₃ (t ₀ ) (15) Formula L ₁ = m ₄ (t ₁ ) −m ₄ (t ₀ ) (16) Formula L ₁ = m ₅ (t ₁ ) −m ₅ (t ₀ ) (17) By the way, depending on the surrounding environment based on the usage pattern of the sensor device Se, the frequency bin 5b including a relatively large background signal (noise) may be known in advance. For example, when there is a device that is supplied with power from a commercial power source around the sensor device Se, the frequency is lower than the commercial power frequency (for example, 60 Hz) (for example, 60 Hz, 120 Hz, etc.) The signal of the frequency bin 5b including the specific frequency is likely to contain a relatively large background signal. On the other hand, the sensor signal output from the radio wave sensor 1 when the object to be detected is moving in the detection area has a frequency (Doppler frequency) of the sensor signal, the distance between the radio wave sensor 1 and the object, Since it changes at any time according to the moving speed of the object, it does not occur constantly at a specific frequency.

Therefore, when each filter bank 5a has a plurality of frequency bins 5b, the signal processing device 2 sets the frequency bin 5b in which the background signal is steadily included as the specific frequency bin 5b _i, and removes the background signal. 10 invalidates the signal of the specific frequency bin 5b _i in the first mode and complements it with the signal strength estimated from the signal strength of the two frequency bins 5b adjacent to the specific frequency bin 5b _i. The signal may be removed. In the examples of FIGS. 6A and 6B, the third frequency bin 5b from the left in FIG. 6A is the specific frequency bin 5b _i , and the signal of the specific frequency bin 5b _i (signal strength b) ₃₎ and disable, as shown in FIG. 6 (b), in the particular frequency bin 5b intensity of _i signal components of the two frequency bins 5b proximate to b _2, b ₄ intensity of the estimated signal component from b3 Complement. In this estimation, the average value of the signal strengths b ₂ and b ₄ of the _two frequency bins 5b adjacent to the specific frequency bin 5b _i , that is, (b ₂ + b ₄ ) / 2 is used as the estimated signal strength. It is set to b _3. In short, i-th frequency bin 5b from the low-frequency side in the filter bank 5a are identified frequency bin 5b _i, if the intensity of the signal of the specific frequency bin 5b _i and b _i, b _i is the following ( The value obtained by the estimation equation shown in equation 18) is used.
b _i = (b _i−1 + b _{i + 1} ) / 2 (18) Thereby, the signal processing device 2 can reduce the influence of the background signal (noise) of the specific frequency that is constantly generated in a shorter time. Is possible. Therefore, the signal processing device 2 can improve the detection accuracy of the detection target object.

  The first recognizing means 7a performs recognition processing for identifying an object based on the distribution in the frequency domain of each normalized intensity that has passed through each filter bank 5a and is normalized by the normalizing means 6. Here, the identification is a concept including classification and recognition.

  The first recognition means 7a can identify an object by performing pattern recognition processing based on principal component analysis, for example. The first recognition means 7a operates according to a recognition algorithm using principal component analysis. In order to employ such first recognizing means 7a, learning sample data in the case where the detection area of the radio wave sensor 1 does not include an object to be detected and learning samples corresponding to different movements of the object to be detected in advance. Data is obtained, and data obtained by performing principal component analysis on the plurality of learning data is stored in the database 11. Data stored in the database 11 is data used for pattern recognition, and is category data in which an object motion is associated with a projection vector and a discrimination boundary value (threshold value).

Here, for convenience of explanation, the distribution in the frequency domain of the normalized intensity corresponding to the learning sample data when the detection object of the radio wave sensor 1 does not include the detection target object is shown in FIG. It is assumed that the distribution in the frequency domain of the normalized intensity corresponding to the learning sample data when it is included is FIG. In FIG. 7A, the normalized intensities of the signals that have passed through the respective filter banks 5a are m ₁₀ , m ₂₀ , m ₃₀ , m _40, and m _{50 in} order from the low frequency side. In FIG. 7B, the normalized intensities of the signals that have passed through the respective filter banks 5a are m ₁₁ , m ₂₁ , m ₃₁ , m _41, and m _{51 in} order from the low frequency side. Then, FIG. 7 (a), the in any of (b), the sum of the normalized intensity of the signal passing through each of the three filterbanks 5a of the low frequency side and variables m _1, 2 two filter banks 5a of the high-frequency side Let the variable m _{2 be} the sum of the normalized intensities of the signals that have passed through each. In short, in FIG. 7A, the variables m ₁ and m ₂ are obtained by the following equations (19) and (20), respectively.
m ₁ = m ₁₀ + m ₂₀ + m ₃₀ (19) Formula m ₂ = m ₄₀ + m ₅₀ (20) In FIG. 8B, the variables m ₁ and m ₂ are the following formulas (21) and (22) ) Is determined by each formula.
m ₁ = m ₁₁ + m ₂₁ + m ₃₁ (21) Equation m ₂ = m ₄₁ + m ₅₁ (22) Equation 7 (c) shows two dimensions when the two variables m ₁ and m ₂ are orthogonal to each other. The scatter diagram, the projection axis, and the identification boundary are illustrated two-dimensionally in order to explain in an image. In FIG. 7C, the coordinate position of each scatter point (“+” in FIG. 7C) in the region surrounded by the broken line is μ0 (m ₂ , m ₁ ), and each of the points in the region surrounded by the solid line. The coordinate position of the scattering point is μ1 (m ₂ , m ₁ ). In the principal component analysis, a group Gr0 of data corresponding to learning sample data when the detection area of the radio wave sensor 1 does not include an object to be detected and an object to be detected are included in the detection area of the radio wave sensor 1 in advance. A data group Gr1 corresponding to the learning sample data is determined. In the principal component analysis, the distribution of data (schematically shown by broken lines and solid lines) in which the scattered points in the respective regions surrounded by the broken lines and solid lines in FIG. 7C are projected on the projection axis. The projection axis is determined under the condition that the average interval is maximized and the variance is maximized. Thereby, in principal component analysis, a projection vector can be obtained for each learning sample.

  By the way, it is preferable that the signal processing apparatus 2 includes an output unit 12 that outputs the identification result obtained by the first recognition unit 7a. Then, for example, when the object to be detected is recognized by the first recognition unit 7a, the signal processing device 2 outputs an output signal ("1") indicating that the object has been detected from the output unit 12 to detect the object. When the target object is not recognized, it is preferable that the output unit 12 output an output signal (“0”) indicating that the object is not detected.

  In FIG. 1, the signal processing device 2 is realized by executing an appropriate program by a microcomputer except for the amplification unit 3, the A / D conversion unit 4, the output unit 12, and the database 11.

  Here, a relationship between an example of a sensor signal output from the radio wave sensor 1 and an output signal output from the output unit 12 will be described with reference to FIGS.

  FIG. 8 illustrates the usage state of the sensor device Se, and shows that the object Ob to be detected is a person, and a tree Tr that is an object other than the detection target exists in the outdoor detection area. ing. FIG. 9 shows the radio wave sensor 1 when the object Ob moves by 6.7 m at a moving speed of 1 m / s in front of the tree Tr with the branches and leaves of the tree Tr swaying under this usage condition. An example of the output sensor signal is shown. The distance between the radio wave sensor 1 and the tree Tr is about 10 m, and the distance between the radio wave sensor 1 and the object Ob is about 8 m. FIG. 10 is a diagram showing a distribution of normalized strength in the frequency domain and a distribution in the time axis domain. FIG. 11 is an output signal of the output unit 12, and it has been confirmed that false detection caused by the movement of an object other than the detection target can be reduced.

  By the way, if you look at the distribution of normalized intensity in the frequency domain, if the object in the detection area is a tree, it will not move even if branches and leaves shake, so the object will move in the detection area. Compared to the case of a walking person, the signal component has a frequency distribution from a lower frequency region. On the other hand, if the object is a person walking in the detection area, it has a mountain-shaped frequency distribution with a center frequency near the frequency corresponding to the walking speed, and there is a clear difference in the frequency distribution. .

  Objects other than the detection target existing in the detection area are mainly movable objects that are not moving objects. When the detection area of the radio wave sensor 1 is set outdoors, the object other than the detection target existing in the detection area is not limited to the tree Tr, and examples thereof include an electric wire that is swayed by the wind.

  Here, the relationship between another example of the sensor signal output from the radio wave sensor 1 and the output signal output from the output unit 12 will be described with reference to FIGS.

  FIG. 12 illustrates the usage state of the sensor device Se, and shows that the object Ob to be detected is a person and it is raining in the outdoor detection area. FIG. 13 shows an example of a sensor signal output from the radio wave sensor 1 when the object Ob moves by 6.7 m at a moving speed of 1 m / s under this usage condition. FIG. 14 is an output signal of the output unit 12 when the background signal removal unit 10 does not remove the background signal in the first mode. FIG. 15 is a diagram showing the distribution of the normalized intensity in the frequency domain and the distribution in the time axis domain when the background signal is removed by the background signal removing unit 10 in the first mode. FIG. 16 is an output signal of the output unit 12 when the background signal is removed by the background signal removing unit 10 in the first mode. In FIG. 16, when the output signal is 0, “*” indicating a correct result is surrounded by white circles (◯). From the comparison between FIG. 16 and FIG. 14, in the sensor device Se, the background signal is removed by the background signal removing unit 10 in the first mode, thereby causing the movement of an object other than the detection target (here, raindrops). It can be seen that false detection is reduced.

  Further, when the detection area of the radio wave sensor 1 is set indoors, an object other than the detection target existing in the detection area is, for example, a device including a movable body (a blade in the case of a fan) such as a fan. Is mentioned.

  The signal processing device 2 preferably makes the above-described threshold variable by setting from the outside. As a result, the signal processing device 2 can adjust the false alarm rate and false alarm rate required according to the intended use. For example, in an application where the object to be detected is a person and the lighting load is controlled to be turned on / off based on an output signal from the output unit 12, the person enters the detection area of the radio wave sensor 1. Some misinformation may be tolerated as long as it is misreported.

  As described above, the signal processing apparatus 2 includes the amplification unit 3, the A / D conversion unit 4, the frequency analysis unit 5, the normalization unit 6, and the first recognition unit 7a. Here, the frequency analysis means 5 converts the sensor signal output from the A / D converter 4 into a signal in the frequency domain, and extracts it as a signal for each filter bank 5a in a group of filter banks 5a having different frequency bands. Further, the normalization means 6 standardizes the intensity of the signal that has passed through each filter bank 5a by the sum of the signals extracted by the frequency analysis means 5 or the sum of the intensities of the signals that have passed through the predetermined plurality of filter banks 5a. And output as normalized strength. The first recognizing unit 7 a performs a recognition process for identifying an object based on a frequency distribution determined from the normalized intensity for each filter bank 5 a output from the normalizing unit 6. As a result, the sensor device Se can reduce false detection caused by the movement of an object other than the detection target. In short, in the sensor device Se, it is possible to separate and identify an object having a statistically different frequency distribution determined by a normalized intensity obtained by standardizing signals that have passed through a plurality of predetermined filter banks 5a, thereby reducing false detection. It becomes possible to do.

  Further, in the filter bank 5a using FFT, a process of multiplying a sensor signal by a predetermined window function (window function) is performed before the FFT process, and a side lobe outside a desired frequency band (pass band) ( Side-lobe) may need to be suppressed. As the window function, for example, a rectangular window, a Gauss window, a Hann window, a Hamming window, or the like can be used. On the other hand, in the filter bank 5a using DCT, the window function can be eliminated or the window function can be realized with a simple digital filter.

  Also, the filter bank 5a using DCT is compared with the filter bank 5a using FFT, whereas FFT is a complex arithmetic processing method (calculating intensity and phase), whereas DCT is a real arithmetic processing method. Therefore, it is possible to reduce the operation scale. In addition, in the DCT, when compared with the FFT with the same number of processing points, the frequency resolution is ½ that of the FFT. Therefore, it is possible to reduce the hardware resources such as the database 11 and the like. Become. In the signal processing device 2, for example, when the number of samplings per second of the A / D conversion unit 4 is 128 (when the sampling frequency is 1 kHz), the width of the FFT bin 5b is 8 Hz, The width of the DCT bin 5b can be 4 Hz. These numerical values are merely examples, and are not particularly limited.

  Further, when the object to be detected is continuously identified on the time axis by the first recognition unit 7a, the signal processing device 2 uses the normalized strength output from the normalization unit 6 at that time as the background of the offset. By removing it as a signal, the recognition accuracy can be improved.

  The first recognizing unit 7a is not limited to identifying an object by pattern recognition processing by principal component analysis, and may identify an object by pattern recognition processing by KL conversion, for example. The signal processing device 2 reduces the calculation amount in the first recognition unit 7a and stores the database 11 by performing pattern recognition processing by principal component analysis or pattern recognition processing by KL conversion in the first recognition unit 7a. The capacity can be reduced.

  The first recognizing unit 7a may perform a recognition process for identifying an object based on the component ratio of the normalized intensity for each filter bank 5a output from the normalizing unit 6.

  Such first recognition means 7a may be configured to identify an object by performing recognition processing by multiple regression analysis, for example. In this case, the 1st recognition means 7a operate | moves according to the recognition algorithm using multiple regression analysis.

When such first recognition means 7a is employed, learning data corresponding to different movements of the object to be detected within the detection area of the radio wave sensor 1 is acquired in advance, and the plurality of learning data is obtained. Data obtained by performing multiple regression analysis is stored in the database 11. According to the multiple regression analysis, as shown in FIG. 17, the synthesized waveform Gs obtained by synthesizing the signal component s1 ′, the signal component s2 ′, and the signal component s3 ′ is the type of the signal components s1, s2, and s3, and the signal component. Even if the intensity of each of the signal components s1, s2, and s3 is unknown, it is possible to separate and estimate the signal components s1, s2, and s3 from the combined waveform Gs. [S] in FIG. 17 indicates a matrix having signal components s1, s2, and s3 as matrix elements. [S] ⁻¹ in FIG. 17 means an inverse matrix of [S]. 17 in FIG. 17 means a matrix having the normalized intensity component ratios (coefficients) i1, i2, and i3 as matrix elements. Here, the data stored in the database 11 is data used for the recognition process, and is data in which the motion of the object is associated with the signal components s1, s2, and s3.

  In FIG. 18A, the horizontal axis represents time, the vertical axis represents normalized strength, and a person who is an object to be detected is 10 m at a moving speed of 2 m / s under an oscillating electric wire in an outdoor detection area. The data (corresponding to the above-mentioned synthesized waveform Gs) of the normalized strength output from the normalizing means 6 when it has moved only by A1 is indicated as A1. FIG. 18A also shows signal components A2 and A3 separated from data A1 by multiple regression analysis. Here, the signal component A2 is a signal component resulting from movement of a person, and the signal component A3 is a signal component resulting from shaking of the electric wire. FIG. 18B shows that the recognition means 7 recognizes that an object to be detected exists when A2> A3, and sets the output signal of the output unit 12 to “1”. Otherwise, the object to be detected exists. This is an output signal of the output unit 12 when it is identified that the output signal of the output unit 12 is “0”. From FIG. 18B, it was confirmed that false detection caused by the movement of an object other than the detection target (here, an electric wire) can be reduced.

  The signal processing apparatus 2 preferably makes the above-described determination condition (A2> A3) variable by setting from the outside. For example, it is preferable that the determination condition is A2> α × A3, and the coefficient α is variable by setting from the outside. As a result, the signal processing device 2 can adjust the false alarm rate and false alarm rate required according to the intended use.

  Note that the first recognition unit 7a may identify an object to be detected based on the characteristics of the frequency distribution and the component ratio of the normalized intensity.

  The first recognizing means 7a may identify an object by majority decision based on the result of odd number of recognition processes on the time axis. For example, according to the majority decision as a result of the recognition processing performed three times in the region surrounded by the one-dot chain line in FIG. 19, the output signal of the output unit 12 is “1”.

  Thereby, the signal processing apparatus 2 can improve the identification accuracy in the 1st recognition means 7a.

Further, the signal processing device 2 recognizes the first recognition means 7a only when the sum of the signal component intensities or weighted sums of the plurality of predetermined filter banks 5a before normalization by the normalization means 6 is not less than a predetermined value. Processing may be performed or the recognition result by the first recognition means 7a may be validated. FIG. 20 shows an example in which the signal strength of each filter bank 5a before normalization by the normalizing means 6 is m ₁ , m ₂ , m ₃ , m ₄ and m _{5 in} order from the low frequency side. FIG. 20A shows a case where the sum of the intensities (m ₁ + m ₂ + m ₃ + m ₄ + m ₅ ) is equal to or greater than a predetermined value (Eth), and FIG. 20B shows the sum of the intensities (m ₁ + m ₂ + m _3). + M ₄ + m ₅ ) is less than a predetermined value (Eth).

  Thereby, the sensor device Se can reduce erroneous detection.

  The first recognition means 7a may have a function of identifying an object by performing recognition processing using a neural network as recognition processing. Thereby, the signal processing apparatus 2 can improve the identification accuracy by the first recognition means 7a.

  As the neural network, for example, an unsupervised competitive learning type neural network can be used. As a neural network, a super propagation back propagation type can be used, but a competitive learning type neural network is simpler than a back propagation type neural network, It is only necessary to learn using the learning data for each category, and it is possible to reinforce learning by additional learning even after learning once.

  As the learning data, the output of the normalizing means 6 acquired in advance corresponding to each different movement of the object to be detected in the detection area of the radio wave sensor 1 is used.

  For example, as shown in FIG. 21, the neural network includes two layers of an input layer 71 and an output layer 72, and each neuron N2 of the output layer 72 is coupled to all the neurons N1 of the input layer 71. Have. The neural network is assumed to be realized by executing an appropriate application program with a microcomputer, but a dedicated neurocomputer can also be used.

  The operation of the neural network includes a learning mode and a detection mode. After learning using appropriate learning data in the learning mode, recognition processing is performed in the detection mode.

  The degree of connection (weight coefficient) between the neuron N1 in the input layer 71 and the neuron N2 in the output layer 72 is variable. In the learning mode, the learning data is input to the neural network to learn the neural network. The weighting coefficient between each neuron N1 and each neuron N2 of the output layer 72 is determined. In other words, each neuron N2 in the output layer 72 is associated with a weight vector whose element is a weight coefficient between each neuron N1 in the input layer 71. Therefore, the weight vector has the same number of elements as the neuron N1 of the input layer 71, and the number of feature parameters input to the input layer 71 matches the number of elements of the weight vector.

  On the other hand, in the detection mode, when the detection data output from the normalizing means 6 for determining the category is given to the input layer 71 of the neural network, the weight vector and the detection data among the neurons N2 of the output layer 72 The neuron N2 having the smallest distance is fired. If a category is associated with the neuron N2 of the output layer 72 in the learning mode, the category of the detection data can be known from the category of the position of the fired neuron N2. In the neural network in the sensor device Se of the present embodiment, the category is set so that everything except “detection (detection)” of the object is determined as “non-detection (non-detection)”.

  By the way, as described above, the signal processing device 2 includes the second recognition unit 7b that operates when the object to be detected is identified by the first recognition unit 7a and identifies a stationary background signal. The stationary background signal is a background signal that is continuously generated for a predetermined time (for example, 3 seconds). The stationary background signal is, for example, stationary noise caused by the operation of a device (for example, a lighting fixture or a fan) in the vicinity of the sensor device Se, rain, or the like.

  When the first recognizing means 7a identifies (detects) the object (yes in process St1 in FIG. 22), the second recognizing means 7b confirms the presence or absence of a steady background signal (background) (FIG. 22). Process St2) When it is recognized that a stationary background signal exists, the stationary background signal is identified (yes in process St2 of FIG. 22). Then, when the stationary background signal is identified by the second recognizing unit 7b, the sensor device Se changes the background signal to the background signal identified by the second recognizing unit 7b (processing St3 in FIG. 22). Transition to the second mode (process St1 in FIG. 22). In the second mode, after the background signal is removed by the background signal removing means 10, the recognition processing by the first recognition means 7a is performed. Therefore, the sensor device Se can efficiently remove a stationary background signal, and can reduce erroneous detection due to the movement of an object other than the detection target. Further, the sensor device Se confirms the presence or absence of a stationary background signal (background), and when the stationary background signal is not recognized (no in process St2 in FIG. 22), performs a recognition process by the first recognition means 7a ( Process St4 in FIG. 22). If the object is not identified (detected) (no in process St4 in FIG. 22), the process returns to process St1. In FIG. 22, the process St1 is a process in a state of waiting for detection of an object to be detected, whereas the process St4 is a process in a state after detection of the object to be detected.

  In the sensor device Se, it is preferable that the second recognizing unit 7b periodically recognizes a steady background signal and updates (changes / cancels) the background signal (St5 and St6 in FIG. 22). Thereby, the sensor device Se suppresses the background signal removing unit 10 from using the previous background signal when the intensity of the stationary background signal changes or when the stationary background signal disappears. This makes it possible to further reduce false detections.

  The second recognizing means 7b outputs the background signal based on at least one of the characteristics of the frequency distribution determined from the normalized strength for each filter bank 5a output from the normalizing means 6 and the characteristics of temporal continuity of the frequency distribution. Identify.

  The characteristic of the frequency distribution is an instantaneous frequency distribution, and means, for example, a distribution in the frequency domain (on the frequency axis) of the normalized intensity at a certain time as shown in FIG. The temporal continuity of the frequency distribution means stationarity and means statistical stability.

  When identifying the background signal based on the characteristics of the frequency distribution, the second recognizing unit 7b performs component analysis of the characteristics of the frequency distribution to identify the background signal. In the component analysis, whether or not the characteristics of the frequency distribution are realized by any one or a combination of a plurality of frequency distribution components that are stored in the memory as a reference of the background signal, and the background signal is identified based on the result. To do. In this case, the second recognizing unit 7b may perform a recognition process by multiple regression analysis, for example. The frequency distribution component means a distribution in the frequency domain of normalized intensity.

  For example, when the background signal is identified based on the temporal continuity characteristics of the frequency distribution, the second recognizing unit 7b uses the same filter as the pulse-like signal that exceeds the specified value (for example, 0.1) in the normalized intensity. If the bank 5a has a predetermined number of times (for example, 10 times) or more in a predetermined time (for example, about 3 seconds), the average value of the pulse-like signal for each filter bank 5a or a constant to the average value Is recognized as a stationary background signal. FIG. 24 shows an example of the frequency distribution in the case where the pulse-shaped signal is less than the specified number of times in a predetermined time. FIG. 25 shows an example of the frequency distribution in the case where the pulsed signal is equal to or more than the specified number of times in a predetermined time. FIG. 26 shows an example of an instantaneous frequency distribution including a stationary background signal.

  Further, when the second recognizing means 7b identifies the background signal based on the temporal continuity characteristics of the frequency distribution, the signal of each filter bank 5a is flat when the normalized intensity exceeds a specified value (for example, 0.1). Average frequency for each filter bank 5a or a constant multiplied by a constant when there is a certain frequency distribution over a specified number of times (for example, 10 times) during a predetermined time (for example, about 3 seconds) Is recognized as a stationary background signal. The second recognizing means 7b is realized by executing an appropriate program with the microcomputer.

Se sensor device 1 Radio wave sensor 2 Signal processing device 3 Amplification unit 4 A / D conversion unit 5 Frequency analysis unit 5a Filter bank 5b Frequency bin 5b _i Specific frequency bin 6 Normalization unit 7a First recognition unit 7b Second recognition unit 8 Average Processing means 10 Background signal removing means

Claims (11)

  A radio wave sensor that transmits a radio wave and receives a radio wave reflected by an object; and a signal processing device that performs signal processing on a sensor signal corresponding to the movement of the object output from the radio wave sensor. From the amplification unit that amplifies the sensor signal output from the radio wave sensor, the A / D conversion unit that converts the sensor signal amplified by the amplification unit into a digital sensor signal, and the A / D conversion unit A frequency analysis means for converting the output sensor signal into a signal in the frequency domain and extracting it as a signal for each filter bank in a group of filter banks having different frequency bands; and a sum of signals extracted by the frequency analysis means or a predetermined value The sum of the strengths of the signals that have passed through the plurality of filter banks is normalized to regulate the strength of the signals that have passed through each of the filter banks. A recognition process for identifying the object by at least one of a normalization unit that outputs the normalized intensity and a frequency distribution determined from the normalized intensity for each filter bank output from the normalizing unit or a component ratio of the normalized intensity A first recognition unit that performs the above operation, a background signal removal unit that removes a background signal from the signal that has passed through the filter bank, and an operation that is performed when the object is identified by the first recognition unit, and is output from the normalization unit Second recognition means for identifying a stationary background signal by at least one of a characteristic of a frequency distribution determined from a normalized intensity for each filter bank and a characteristic of temporal continuity of the frequency distribution. A sensor device comprising:   The sensor device according to claim 1, wherein the second recognizing unit periodically recognizes a stationary background signal and updates the background signal.   The frequency analysis means has a function of converting the sensor signal output from the A / D converter into a signal in the frequency domain by performing discrete cosine transform, and each of the filter banks has a plurality of frequency bins. And the signal processing device includes a smoothing processing unit between the frequency analysis unit and the normalization unit, and the smoothing processing unit is configured to output a signal for each frequency bin for each filter bank. The apparatus has at least one of a function of smoothing the intensity in the frequency domain and a function of smoothing the intensity of the signal for each frequency bin in the time axis direction for each filter bank. 3. The sensor device according to 1 or 2.   4. The sensor device according to claim 1, wherein the background signal removing unit removes the background signal by subtracting the background signal from a signal that has passed through the filter bank. 5.   5. The sensor device according to claim 4, wherein the background signal removing unit removes, as the background signal, an average value on a time axis of a plurality of signals for each filter bank obtained in advance.   Each of the filter banks has a plurality of frequency bins, the signal processing apparatus sets the frequency bin in which the background signal is constantly included as a specific frequency bin, and the background signal removal unit includes the specific frequency bin. 6. The background signal is removed by invalidating the signal and supplementing with the signal strength estimated from the signal strength of the frequency bin adjacent to the specific frequency bin. The sensor device according to item.   The sensor device according to claim 1, wherein the first recognition unit identifies the object by performing pattern recognition by principal component analysis or KL conversion as the recognition processing.   The sensor device according to claim 1, wherein the first recognition unit identifies the object by performing recognition processing by multiple regression analysis as the recognition processing.   The sensor device according to claim 1, wherein the first recognition unit identifies the object by performing recognition processing using a neural network as the recognition processing.   10. The sensor device according to claim 1, wherein the first recognizing unit determines an identification result by a majority decision based on an odd number of recognition processing results on a time axis. .   The signal processing apparatus performs the recognition processing by the first recognition unit or the first recognition unit only when the sum of the intensities of the signals before standardization by the standardization unit is a predetermined value or more. The sensor device according to claim 1, wherein the result of the recognition process is validated.