JP5526363B2 - Far-infrared buried object automatic detection device - Google Patents

Description

  The present invention converts the temperature difference due to the far-infrared radiation intensity caused by the temperature difference between the buried object and the ground surface into the luminance data of the far-infrared camera, and the far-infrared embedding that roughly determines the location of the buried object from the luminance distribution The present invention relates to an automatic object detection apparatus.

  Disperse denitrifying bacteria, which are microorganisms that convert nitrogen oxides, which are the explosive composition material that forms the explosives of obstacles buried in the ground, into nitrogen gas, and hydrocarbon-assimilating bacteria that decompose hydrocarbons on the ground surface Patent Document 1 below discloses a method for detecting whether or not an obstacle is buried by detecting the temperature distribution on the surface of the earth whose temperature has changed due to the presence of the microorganisms after a predetermined time has passed by an infrared imaging camera and an image display device. Yes.

  A guide rail mounted on the vehicle, a carriage that can move along the guide rail, a microwave irradiation device that is mounted on the carriage and irradiates the ground with microwaves, and the microwave irradiation device are juxtaposed with each other. An infrared camera that is mounted on the carriage and receives infrared rays from the ground surface, and an infrared imaging device for identifying an embedded obstacle provided in the control room of the vehicle to receive signals from the infrared camera and display infrared images This is an embedded obstacle exploration device that is configured to provide an obstacle exploration device for exploring obstacles buried in the ground by non-destructive means. Japanese Patent Application Laid-Open No. 2004-228688 discloses a method for visualizing the behavior of temperature so that the presence or absence of an embedded obstacle can be determined regardless of the material of the embedded obstacle.

  A transient temperature difference due to the difference in specific heat between the buried obstacle and the ground surface is generated by spraying the cooling liquid on the ground surface where there may be a buried obstacle and forcibly cooling or heating with a heating burner. The embedded obstacle from the position map display luminance image of the embedded obstacle composed of the laser marker installed in the detection area, the infrared camera that can measure the detection area, the image signal, the laser distance measuring device and the self-positioning device A method for detecting an object is disclosed in Patent Document 3 below.

  As a method for determining the presence or absence of buried obstacles to enable remote sensing of the presence or absence of buried obstacles using an infrared sensing method, a simulated buried obstacle with a rubber plate placed on the upper surface of a plastic part with a built-in explosive part , Buried in the surrounding simulated soil body, built a 3D physical model in the soil for 3D numerical simulation, the ground surface of the surrounding simulated soil body in the 3D physical model as the heating surface and heat dissipation surface, A temperature change over 24 hours due to solar radiation is reproduced by a three-dimensional numerical simulation performed using a non-stationary three-dimensional heat conduction equation as a basic equation under a predetermined initial condition and environmental conditions, and obtained by a reproduction result. Determine the optimum time zone determined by the relationship with the clear image of the buried obstacle obtained from each two-dimensional thermal image for each elapsed time, and determine the optimum time zone as an empirical rule How to enable infrared remote sensing of the presence or absence of actual buried obstacles in the natural environment and is disclosed in Patent Document 4.

  The heat generation state of the printed wiring board on which the electronic component is mounted is captured by a thermal image, and abnormality or deterioration of the printed wiring board is diagnosed by fractal dimension analysis based on the temporal change of the thermal image. When detecting deterioration of a printed wiring board, the fractal dimension is calculated for the thermal image at each observation, and the fractal dimension judgment value in the failure state of the printed wiring board is obtained in advance from the temporal change tendency of this fractal dimension. A method for determining deterioration by comparing with the above is disclosed in Patent Document 5 below.

JP 2004-132634 A JP-A-6-235599 JP 2001-74397 A JP 2005-195287 A JP 2001-337059 A

  In the case of Patent Document 1, the target detection is facilitated by converting or decomposing the explosive composition material of the obstacle buried in the ground and generating or enhancing the contrast of the temperature distribution of the target and the ground surface, It is necessary to prepare the microorganisms to be sprayed and the amount to be sprayed, which not only increases the size and cost of the device, but also requires time to change the temperature, so it is difficult to detect the buried obstacle in real time while the vehicle is running .

  In the case of Patent Document 2, it is possible to deal with a target remotely by using a microwave irradiation device that irradiates the ground surface with microwaves. However, even when using a frequency with good underground permeability, the vehicle travels in real time. In order to increase the target temperature separated by 7 m or more, a considerable transmission output is required, and the apparatus becomes large.

  Furthermore, the method of Patent Document 3 in which a coolant is sprayed on the ground surface and heated by a forced cooling or heating burner is relatively difficult to generate a transient temperature difference due to the difference in specific heat between the buried obstacle and the ground surface. Although it is fast but difficult to process in real time, there is a limit to the traveling speed even if it is mounted on the vehicle, and it is difficult to significantly improve work efficiency.

  On the other hand, as a method for determining the presence or absence of buried obstacles so that the remote sensing of the presence or absence of buried obstacles can be performed using an infrared sensing technique, a simulated buried obstacle is buried in the surrounding simulated soil body and three-dimensional physics in the soil. The model is constructed, the ground surface is the heating surface and the heat dissipation surface, and the temperature change over 24 hours due to solar radiation is used as the basic equation using the unsteady three-dimensional heat conduction equation under the predetermined initial conditions and environmental conditions. The method of Patent Document 4, which is reproduced by a three-dimensional numerical simulation performed and compared with a database from a two-dimensional thermal image of an infrared camera image, has an extremely large number of problems such as soil type, moisture content, existence of vegetation, season, etc. There is a need to have a database of deterministic conditions, and this approach requires a sophisticated analysis device instead of requiring a large device other than an infrared camera.

  By the way, there is a method of Patent Document 5 that performs simple feature extraction using an fractal dimension from an image obtained by an infrared camera. It is a technique to capture the heat generation state of the printed wiring board with a thermal image, analyze the temporal change of the fractal dimension, and diagnose abnormality and deterioration of the printed wiring board, but it has a long-term database on failure, In buried object detection like the present invention, since it is used in an unknown environment, it is difficult to suppress false alarms by using the fractal dimension itself as a feature quantity for target extraction. This method is undecided and is not suitable for other applications.

  In addition, a general method for identifying a two-dimensional image by a neural network or shape pattern matching may improve the performance due to a learning effect, but the image conversion, filter processing, clustering, and threshold processing are performed on the two-dimensional image in advance. Because it requires high-speed and complicated parallel processing such as identification after removing noise and false targets by screening, etc., it is fast for real-time buried obstacle detection while driving the vehicle Since a large-capacity computer is required, it is not worth the cost.

  Therefore, real-time data with a higher detection probability can be applied to a wide range of environmental parameters with robustness by simple data processing from data acquired by a simple device such as an on-board infrared camera. The challenge is to produce a target detection technique and apparatus that is inexpensive to process. The key to enable this purpose is the CFAR (Constant False Alarm Rate) technology for infrared detection. There are still few studies on the properties of ground surface clutter in the far infrared, so the clutter statistics are not clear. However, since the CFAR cannot exhibit high performance except for the parametric CFAR that assumes specific statistical properties, development of a clutter suppression technique peculiar to an infrared image becomes a problem.

  The present invention has been made in recognition of such a situation, and the purpose of the present invention is to provide a far infrared ray with a simple configuration even in an environment where target signal extraction from a buried object by a far infrared camera is hindered by ground surface clutter. Provided is a far-infrared buried object automatic detection device that can suppress ground surface clutter in a region and extract a target signal in real time, and thus can detect a buried object while a vehicle equipped with a far-infrared camera is moving away. There is.

In order to achieve the above object, a far-infrared buried object automatic detection device according to an aspect of the present invention is configured to automatically detect a target buried in the ground by mounting a far-infrared camera on a traveling vehicle in real time. ,
From the time-series data of one-dimensional luminance in the azimuth direction in the depression angle range limited by the far-infrared camera, it is complicated using the emissivity, radiation temperature and emissivity coefficient of all factors including the target and the ground clutter. It is characterized by modeling by the fractional Brownian motion without solving the heat conduction equation under the distributed boundary conditions and simulating the phenomenon related to the radiation intensity distribution in the far-infrared region of ground clutter including vegetation .

  In the above aspect, a code discrimination function is incorporated into the CA-CFAR (Cell Averaged CFAR, hereinafter referred to as CA-CFAR) process as a CFAR process adapted to the statistical properties of the ground surface clutter in the far infrared region. A clutter suppression process may be performed.

  In the above aspect, by converting the time series data of the one-dimensional luminance to fractional differentiation, the colored noise that is the ground surface clutter unique to the far-infrared region is converted into uncorrelated white normal noise, and the converted time series data It is preferable to take the difference between the output average and the time-series data after conversion, square the result, and then perform the CA-CFAR process.

  In the above aspect, the negative jump point where the negative value is obtained by taking the difference between the average of the time-series data converted to the uncorrelated white normal noise and the time-series data after conversion, and the absolute value is maximized. However, the amplitude intensity distribution of the clutter may be adapted to the CA-CFAR process by squaring the result.

  In the above aspect, the sign of the data immediately before squaring the difference between the average of the time-series data converted to the uncorrelated white regular noise and the time-series data after conversion is labeled using a sign function. By identifying, the positive and negative signs of the jumping points are selected from the CA-CFAR processing output, and the target detection of the low or high temperature is switched relative to the ground surface in response to the change in the temperature of day and night. Good.

  In the above aspect, the sign of the data immediately before squaring the difference between the average of the time-series data converted to the uncorrelated white regular noise and the time-series data after conversion is labeled using a sign function. By identifying the positive and negative signs of the jump point from the CA-CFAR processing output, and by performing a logical operation on the encoded output value, both end points of the target signal rise and fall are detected. The target existence section in consideration of the size of the target to be detected may be automatically determined.

  In the above aspect, the original unimodal pulse is converted into a sign-inverted bimodal pulse by fractional differentiation of the one-dimensional luminance time-series data, whereby the target included in the intermediate portion of the target signal The information cancels out apparently, leaving only the rising and falling points of the target signal, and it becomes a section where only the ground clutter exists, thereby reducing or eliminating the influence of mixed targets from the statistics of the ground clutter. CA-CFAR processing may be performed.

  In the above aspect, the target automatic extraction may be supplemented by non-parametric CFAR processing for the one-dimensional luminance time-series data determined to have no fractal property.

  It should be noted that any combination of the above-described constituent elements, and those obtained by converting the expression of the present invention between methods and systems are also effective as aspects of the present invention.

  According to the far-infrared buried object automatic detection apparatus according to the present invention, the ground surface in the far-infrared region can be obtained with a simple configuration even in an environment where target signal extraction from the buried object by the far-infrared camera is hindered by the ground surface clutter. Clutter can be suppressed and the target signal can be extracted in real time. Further, it is possible to detect a buried object while a vehicle equipped with a far infrared camera is running away.

It is a block diagram of one embodiment of the present invention. It is a block diagram of the 1st target detection processing device in an embodiment. It is a block diagram of the 2nd target detection processing apparatus in an embodiment. It is a block diagram of a fractal property judging device in an embodiment. In embodiment, it is a graph which shows the result and fractal property which analyzed the space spectrum after extracting the time series data of the specific depression angle from the brightness | luminance image, and window-processing. In an embodiment, it is a graph which cuts out time series data of each depression from a luminance picture, and shows distribution of fractal dimension for every time. It is a block diagram which shows that the attention cell value in CA-CFAR process is compared with the gamma distribution of 2M degrees of freedom which is the sum of chi square distribution, and it is determined whether it is a target. It is explanatory drawing which divides | segments the logic operation of a target detection logic determination apparatus into five steps. Results of target detection using the first target detection processing device and the second target detection processing device using the same time-series data determined to have fractal characteristics, and target detection using only the second target detection processing device It is a graph which shows a comparison with a result.

  Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the drawings. In addition, the same code | symbol is attached | subjected to the same or equivalent component, member, process, etc. which are shown by each drawing, and the overlapping description is abbreviate | omitted suitably. In addition, the embodiments do not limit the invention but are exemplifications, and all features and combinations thereof described in the embodiments are not necessarily essential to the invention.

  The apparatus shown in the embodiment of the present invention converts the temperature difference due to the far-infrared radiation intensity caused by the temperature difference between the buried object and the ground surface into the brightness by the far-infrared light receiving element of the far-infrared camera, and the brightness distribution. In order to approximate the location of the buried object from the ground, the target signal from the buried object is suppressed in the far-infrared region by suppressing the ground surface clutter in an environment where the target signal is obstructed by the ground surface clutter. By extracting in time, the buried obstacle is detected at a distance while traveling.

  Obstacles and the like embedded in a traveling vehicle are extremely dangerous, and therefore, it is required to detect them as far apart as possible. These underground obstacles can be combined with data using not only an RF sensor but also an IR sensor, thereby improving the detection capability. In particular, since it is effective to improve detection capability with a far-infrared sensor having a relatively low cost, the present invention is limited to a problem by far-infrared remote sensing. In particular, the detection of a target by a small and light far-infrared camera with a wavelength of 8 to 14 μm is sensitive to the slight difference in radiant intensity caused by the difference in specific heat between the target and soil, taking into account changes in sunshine conditions. This has the effect of detecting the target in real time even in an environment where there is a large amount of noise.

  However, the detection of shallow buried targets by radiation intensity using far-infrared rays with a wavelength of 8-14 μm is based on location and time, such as sunshine duration, weather conditions, vegetation type, surface roughness, and moisture content in the ground. The presence of clutter makes it difficult to detect the target because it is strongly affected by heat radiation and absorption locally depending on the target. Under these circumstances, it is extremely difficult to model all factors including targets and clutters for emissivity, radiation temperature, and emissivity coefficient, and to solve the heat conduction equation under complicatedly distributed boundary conditions. Thus, modeling based on Fractional Brownian Motion (hereinafter referred to as FBM) is attracting attention as a means of simulating a set of uncertain and complex phenomena. There are no examples of research on fractal nature of radiation intensity distribution in the far-infrared region of ground clutter. This device converts the far-infrared radiation intensity distribution from the ground surface into the brightness level output of the far-infrared camera, determines that the time-series data has fractal dimension, and conforms to the statistical properties of the clutter in the far-infrared region. We propose a clutter suppression process using the CFAR process. In particular, luminance data obtained from an uncooled far-infrared camera requires a contrivance that requires target detection only from amplitude information because phase information calculated from I and Q components cannot be used like microwaves.

  Therefore, based on the CA-CFAR process, the effectiveness of the target detection process is verified using measured data using a clutter suppression process incorporating a code discrimination function.

  Hereinafter, this will be specifically described with reference to a block diagram. FIG. 1 shows the overall configuration of a far-infrared embedded automatic detection device, which is a far-infrared camera 1, a reference light source 2, a focus adjuster 3, a brightness adjuster 4, an attitude angle control device 5, a video monitor 6, and an initial setting input device. 7, a video recording device 8, a data conversion recording device 9, a window processor 10, a first target detection processing device 11, a second target detection processing device 12, and a target display device 13.

  In the far-infrared buried object automatic detection apparatus of FIG. 1, the angle of depression (aspect angle) of the far-infrared camera 1 mounted on the vehicle is constant in order to improve the safety and speed of detecting the buried target that is the buried obstacle. It is assumed that the far-infrared camera 1 acquires data from a remote position while traveling while keeping the vehicle at a distance (the camera 1 is assumed to be space-stabilized). In natural landscapes, there is a correlation in the azimuth angle direction, but the correlation in the elevation angle direction is weak. Therefore, a time-series data space obtained by one-dimensional scanning of the luminance image of the clutter in the limited depression angle range in the azimuth direction. By obtaining information from the spectrum, the target is detected without performing complicated two-dimensional image correction.

  The initial input setting device 7 has a function that allows the user to manually input the focal length and brightness contrast while watching the video monitor 6. The reference light source 2 is used to calibrate the brightness and temperature with the focus adjuster 3 and the brightness adjuster 4. The initial input setting device 7 outputs set input values of the azimuth angle and the elevation angle of the far-infrared camera 1 to the attitude angle control device 5, and the attitude angle control device 5 uses the elevation / elevation angle sensor and the servo drive device to transmit the far-infrared camera 1. (That is, the depression angle of the camera 1 is kept constant regardless of the change in the attitude of the vehicle). Further, the initial input setting device 7 outputs the window processing range of the image to the first and second target detection processing devices 11 and 12, and is on the road affected by the reflection from the artificial structure existing on the road shoulder by the window processing. The far-infrared radiation is reduced, and the initial setting value of the false alarm probability is output to the first and second target detection processing devices 11 and 12, and the threshold value is controlled. The video recording device 8 records luminance image data in a two-dimensional matrix DAT format. The data conversion recording device 9 cuts out a necessary part of the data of the video recording device 8 and records it in the CSV format. The window processor 10 cuts out the windowed azimuth angle at the specified depression angle, which is the window processing range of the image specified by the initial input setting device 7, and records it as one-dimensional luminance time series data. The time series data output from the window processor 10 is output to the first target detection processing device 11 to perform processing for target detection. Here, the block diagram of the 1st target detection processing apparatus 11 is shown in FIG. The fractal property determination device 14 built in the first target detection processing device 11 determines the fractal property of the time-series data and calculates the fractal dimension. When it is determined that the time-series data is not fractal, the time-series data is transferred to the first target detection processing device 12 and processing for target detection by non-parametric CFAR processing is performed. A block diagram of the second target detection processing device 12 is shown in FIG. A result of the target detected or not detected by the first target detection processing device 11 or the second target detection processing device 12 is output and displayed on the target display device 13.

  The first target detection processing device 11 shown in FIG. 2 includes a fractal determination device 14, a fractional differentiation device 25, a sign converter 26, a chi-square distribution converter 27, a delay device 28, a false alarm calculator 29, and a CA-CFAR process. A device 30, a binarization converter 31, and a target detection logic determination device 32 are provided.

Here, the feature value of the luminance variation of the time series data output from the window processor 10 is regarded as the complexity of the one-dimensional graph. -176, 1997), it is expressed in Hausdorff dimension (hereinafter referred to as Hausdorff dimension), which is one of the fractal dimensions. The fact that a certain fluctuation is a fractal (hereinafter referred to as fractal property) requires that the spectrum P (f) of the time series data be represented in a power form as in equation (1).

(Where f is the spatial frequency)
The fractal dimension is calculated by subjecting time series data to Fourier transform, and substituting the value of the slope of the graph obtained from the log-log graph of the spectrum into equation (2). Here, as described in the literature (Hakki Takayasu: Fractal, Asakura Shoten, March 1993), the relationship of the following equation is established between the spectral inclination β and the fractal dimension D.

The fractal property determination device 14 provides a criterion for calculating a fractal dimension from a spectrum. The judgment conditions are shown below.
Condition 1: 1 <β <3
Here, β is obvious from the equation (2), but indicates the slope of the regression line of the spectrum.
Condition 2: _f <f _c
Here f _c represents the cutoff frequency.

Actually, it is difficult to obtain an accurate fractal dimension such that the spectral characteristic peculiar to a fractal, which is a power in a high spatial frequency region, is buried in white noise due to the accuracy, resolution, S / N degradation and the like of measurement data. Therefore, fractal dimensions are extracted in a limited spatial frequency region where fractal properties are locally established. Fractal determination apparatus method for calculating the cut-off frequency f _c and fractal dimension of 14 shown in FIG. This calculation method is based on the literature (Kan Araki, observation of surface reflection characteristics by fractal dimension using millimeter wave radio altimeter, theory of theory (B), vol.J82-B, no.5, pp.1053-1062, May 1999. .)

  4 includes a spectrum calculator 15, a cutoff frequency detector 16, a cutoff frequency adder 17, a fractal dimension calculator 18, a difference unit 19, an inverse fractal filter 20, a correlator 21, and a white tester. 22, a cut-off frequency determiner 23, and a latch 24.

In order to obtain the parameter d, the fractality determination device 14 performs a closed loop process while correcting the same data in both the frequency and the time domain to improve accuracy. The counterclockwise flow starting from the input time series is to calculate the fractal dimension from the spectrum. Input data Fast Fourier transform spectrum calculator 15 (hereinafter FFT and description) processing the spectral data obtained was then is f _c is sent to the cutoff frequency detector 16 is calculated. Here, determine the slope of the spectrum on a logarithmic scale in the frequency domain of f <f _c, then by the fractal dimension calculator 18 obtains the parameter d from the slope of the straight line. If the parameter converges to a value close to the true value, the fractal dimension D is calculated as a result and output to the latch 24.

  The clockwise flow starting from the input time series is to calculate the fractal dimension from the time domain, and has a path to the differentiator 19, the inverse fractal filter 20, the correlator 21, and the white tester 22.

Next, the relationship between discrete fractal Gaussian noise (hereinafter referred to as dfGn) and discrete white Gaussian noise (hereinafter referred to as WGn) is shown. It is assumed that time series data of dfGn and WGn are {W _d (n)} and {W (n)}, respectively. The transfer function G (Z) from WGn to dfGn can be expressed in equation (3), and further, the series term expansion of equation (3) can be performed, and the relationship between both can be expressed as in equation (4).

Here, Z is a digital quantity representing a time delay, and −0.5 <d <0.5.

Here, k and n are integers, and C (k) is an impulse response.
At this time, if the parameter d equivalent to the fractal dimension cannot be accurately identified, even if the dfGn is passed through the inverse fractal filter 20 with respect to the equation (3), it does not return to the white Gaussian noise, and equivalently (1-Z ⁻¹ ) ⁻ noise and colored of the transfer function of the delta ^d remains. Therefore, by taking the autocorrelation of noise with the correlator 21 and performing the whitening test with the white tester 22, it is possible to confirm whether the value of the fractal dimension is correct. Here, Δd represents an error of the calculated parameter with respect to the convergence value d. If the output is colored noise, it is assumed that the identification of the parameter d has not converged, and the process returns to the counterclockwise flow to change the reading of the spectrum slope β on the logarithmic scale and correct the parameter. Thereafter, the same operation is repeated until the white test result satisfies the desired value, and finally the convergence value d is obtained. In the input data, the true fractal dimension value D is d + 1 because the time series data is subjected to a difference once by the differentiator 19.

 FIG. 5 (a logarithmic value of the spectrum S (f) obtained by the spectrum calculator 15) shows an example of spectrum analysis of time-series data of ground surface reflection clutter in the far infrared region. FIG. 6 shows an example of calculating the fractal dimension at each time (10:00, 12:00, 14:00, 16:00, 18:00).

  If the fractal property is not satisfied, the processing of the first target detection processing device 11 is terminated, and the time series data is transmitted to the second target detection processing device 12.

Assume that the target luminance is a positive number of jump point sets (a finite level square wave signal) expressed by a δ function. To this, a fractional brownian motion (Fractional Brownian Motion, hereinafter referred to as FBM) caused by the radiation from the clutter is added as the luminance of the colored noise. The observed time series data {x (t _k )} is the FBM clutter data sequence {n (t _k )} and the heat to the space due to heat conduction from the embedded target to the ground surface directly above. Assume that it is modeled as the sum of the radiation signal {s (t _k )}. The target is assumed to have thermal radiation, uniform absorption characteristics, and a length in the Az direction. By assuming a set of jump functions having a large value, that is, a square function, even if the target is fractal differentiated with clutter, the target is Since the square wave to be formed is a set of jumping functions, the jumping function here is converted from a unimodal pulse to a bimodal pulse having positive and negative peak values. Will be detected. However, after the fractal differentiation, the square function becomes 0 because the positive and negative values are canceled by the adjacent bimodal pulses in the intermediate section other than the rising and falling end points, and a peak having a large value does not occur. For this reason, the square wave can be treated as an area where the signal level is 0 and only the clutter exists. The CFAR process has the advantage of preventing misidentification of targets and clutter. At this time, a section in which the signal {s (t _k )} takes a significant value is detected by the first target detection processing device 11 according to the following procedure. If the fractal determination unit 14 determines that the fractal characteristic is present, the output result of the fractal dimension calculation unit 18 in FIG. 4 is substituted into the equation (5) to calculate the Hurst exponent H, and the time-series data is sent to the fractional differentiator 25 Send it out. Otherwise, the process proceeds to the second target detection processing device 12.

The fractional differentiator 25 converts the cluttered colored noise {n (t _k )} into uncorrelated white normal noise by fractionating the time series data {x (t _K )} using equation (6). To do. However, in calculation, h is not taken as a limit value, but is approximated as a data acquisition interval.

Here, the function Π is a product function, and when p = H + 1/2 and t takes a discrete value, h is a sampling interval.

The local average calculator calculates the sample average μ of the output data sequence {v (t _k )} in which the target is superimposed on the fractionally differentiated and white normalized clutter as shown in Equation (7), and the χ-square distribution conversion The unit 27 squares the difference between the output data sequence {v (t _k )} and its sample average μ, and sends the result to the CA-CFAR processor 30. Signal-to-clutter ratio (Signal to Clutter Ratio: hereinafter referred to as S / C) is low, and it is difficult to detect signals with a simple threshold judgment using a fixed value. Remove clutter.

The output data string {u (t _k )} of the chi-square distribution converter 27 is input to the CA-CFAR processor 30 to obtain an output data string {w (t _k )}. Here, as shown in FIG. 7, the target cell value z is determined to be compared with a gamma distribution having a probability density function of 2 M degrees of freedom shown in Expression (8).

Here, σ represents the variance in the normal distribution, and M represents the number of cells in FIG.
The false alarm probability is expressed by equation (9) if it is a conditional probability when the threshold is y.

Here, f (y) is the value of the probability density function of the threshold value y, and T is the threshold level setting value.

By converting the random variable to y = K ₀ z using the parameter K ₀ that is the initial setting input in the CA-CFAR processor 30, Expression (8) is expressed as Expression (10).

The false alarm probability is obtained as a lower limit value by substituting Equation (10) into Equation (9) and as in Equation (11). The user can be set to a value which the user requests a false alarm probability by adjusting the threshold parameter K _0.

Since the output data string {v (t _k )} is squared before being input to the CA-CFAR processor 30, the negative jump point is detected by the CA-CFAR processor 30 in the same manner as the positive jump point. A false target occurs. Therefore, the code converter 26 substitutes the data string {v (t _k )} into the code function defined by the equation (12), and causes the χ-square distribution converter 27 to restore the storage of the code information before input once again. Here, when it is desired to detect a target having a temperature lower than the ground surface, the output result of the sign function is multiplied by -1. On the other hand, after the binarization converter 31 substitutes the CA-CFAR processor output {w (t _k )} into the binarization function defined in the equation (13), the binarization converter 31 simplifies the information into only target availability information. , The product of the code converter output and the binarized converter output is output as {wb (t _k )} according to the equation (14) and sent to the target detection logic determination device 32.

Here, sgn (x) represents an encoding function.

Here, bin (x) represents a binarization function.

When the value of the output {w _b (t _k )} continuously takes +1, the target detection logic determination device 32 sets only the first t _k to +1 and sets the other values to 0. Further, when the value of {w _b (t _k )} continuously takes −1, only the last t _k is set to −1 and the other values are set to 0. A section {t _k } having a value of {w _b (t _k )} from +1 to −1 is determined as a section where the target exists. However, it is possible to include a value of 0 in the interval from +1 to -1, and the length of the target to be detected is the maximum number of pixels that the target continuously occupies on the line scan. In the case where the intermediate section of 0 that is longer is included, it is determined that the section is not the target even if it is the section from +1 to -1. FIG. 8 shows a logical operation of the target detection logic determination device. The logical operation is composed of five stages described below.

<First stage>
The integrated outputs of the code converter 26 and the binarization converter 31 of the first target detection processing device 11 are shown in the A column. If the positive pulse or negative pulse in row A occurs continuously, only the pulse in the last row (newest time) is left. If it exists in a single pulse, it is left as it is. A positive residual pulse is a B column, and a negative residual pulse is a C column.
Logical operation: B _i = IF (AND (A _i = 1, OR (A _{i + 1} = 0, A _{i + 1} = −1), 1, 0)
C _i = IF (OR (A _i = 0, A _i = −1), A _i , 0)
Where i: time position of the sample (represents the number)

<Second stage>
The B and C row pulses are added to form the D row.
Logical operation: D _i = B _i + C _i

<Third stage>
When the pulse is inverted from positive to negative with respect to the D column, the positive pulse is regarded as the rising edge of the target pulse and a positive pulse is generated in the E column. On the other hand, a negative pulse is regarded as a target pulse drop, and a positive pulse is generated in the F row. However, when the pulse is inverted from positive to negative with respect to the D column, if the interval of 4 or more continues, it is regarded as not the target because the target size limit is exceeded, and the E column and the F column Does not cause a pulse.
Logical operations:
E _i = IF (AND (D _i = 1, OR (D _{i + 1} = -1, D _{i + 2} = -1, D _{i + 3} = -1), 1, 0) )
F _i = IF (AND (D _{i + 3} = −1, OR (D _i = 1, D _{i + 1} = 1, D _{i + 2} = 1), 1, 0) )
However, if i <4, F _i = 0

<Fourth stage>
A positive pulse is generated between the pulses in the E row and the F row to form the G row.
Logical operation: G _i = IF (AND (F _{i + 1} = 0, OR (E _{i + 1} = 1, G _i = 1)), 1, 0)
However, when i = 1, G _i = 0

<Fifth stage>
The E column and the G column are added to obtain an H column. The reason why the F column is not added is that the falling pulse is not included in the true size of the target because the unimodal pulse is originally changed to the bimodal pulse by the fractional differentiation. The target detection logic determination device 32 extracts a section having a value of +1 in the H column as a target existence section and sends the result to the target display device 13.
Logical operation: H _i = E _i + G _i

The time-series data {x (t _k )} determined to have no fractal property by the fractal property determination unit 14 built in the first target detection processing device 11 is subjected to non-parametric CFAR processing by the second target detection processing device 12. To do. A block diagram of the non-parametric CFAR processor is a block diagram of the second target detection processor 12 of FIG. The second target detection processing device 12 includes a target cell 33, a front tapper group 34, a rear tapper group 35, a unit function converter 36, an adder 37, a rank detector 38, and a target detection determiner 39. .

When the non-parametric CFAR processor is used, since the clutter statistics are not assumed to be fractal, Gaussian, Weibull, lognormal distribution, etc., the equation (15) is directly applied to the time series data {x (t _k )}. When a rank value rank (k) is calculated by a rank detector 38 that deserves threshold detection of a nonparametric CFAR processor, and the target detection determiner 39 determines that the value satisfies the equation (16), x (t _k ) is regarded as a target signal s (t _k ), extracted as a target existence section, and sent to the target display device 13.

Here, n represents the number of tappers, and unit () represents a unit function.

Here, K ₀ represents the threshold parameter of the CA-CFAR processor 30 in the equation (9).

  When comparing both CFAR processors, in order to equalize the false alarm probability setting, the number of tappers n of the nonparametric CFAR processor is set to the error set by the first target detection processor 11 using equation (16). It is determined to be equivalent to the alarm probability value.

  An uncooled microbolometer type far-infrared camera is used, the detection wavelength band is 8 to 14 μm, the pixel is 640 × 480, the field angle is 17.2 ° × 12.9 °, the instantaneous visual field is 0.47 mrad, and the NETD is 0 .Infrared radiation intensity is converted to luminance at 1 ° C or lower and output to video recording device and video monitor by NTSC. The far-infrared camera is installed at a height of 1.4m above the ground, the aspect angle is -8 ° from the horizontal, and the target of embedment is a resin and metal cylinder each with a diameter of about 25cm and a height of about 5cm, 10m in front. It is buried in the ground at an embedding depth of 5 cm. The ground surface vegetation (weeds with a height of 5 cm or less) at the burial position that has been dug up is restored to its original state immediately after the target burial. The temperature difference between the background ground surface as the clutter and the target buried ground surface is measured using a reference light source and recorded separately by a data recorder.

  As shown in FIG. 9, the detection probability is improved by about 15% by using the detection by the CA-CFAR processor as compared with the single detection by the non-parametric CFAR processor when the same false alarm rate is set. This was demonstrated using the data.

  According to the present embodiment, the following effects can be achieved.

(1) A far-infrared camera 1 that is space-stabilized is mounted on a traveling vehicle, and a target obstacle embedded in the ground is separated as much as possible and can be automatically detected in real time.

(2) Time series data of luminance obtained by one-dimensional scanning in the azimuth direction (extracting one-dimensional luminance data in the azimuth direction from the two-dimensional image of the camera) in the depression angle range limited by the far-infrared camera 1 Thus, the target can be detected without performing two-dimensional complicated image processing, and the target position can be roughly estimated.

(3) Modeling with FBM without solving the heat conduction equation under complicatedly distributed boundary conditions using far-infrared emissivity, radiation temperature and emissivity coefficients of all factors including targets and ground clutter. It is possible to easily simulate uncertain and complicated phenomena related to the radiation intensity distribution in the far infrared region of the ground surface clutter including vegetation.

(4) Target extraction by suppressing clutter with a simple configuration by having a clutter suppression process that incorporates a code discrimination function into the CA-CFAR process as a CFAR process suitable for the statistical properties of ground clutter in the far-infrared region Is possible.

(5) By converting the time-series data obtained by one-dimensional scanning to fractional differentiation, the colored noise, which is the ground surface clutter unique to the far-infrared region, is converted into uncorrelated white normal noise, and the output average of the time-series data The CA-CFAR process can be applied by taking the difference from the time-series data after conversion and squaring the result.

(6) The negative jump point where the absolute value becomes maximum by taking the difference between the average of the time-series data converted by fractional differentiation and the time-series data after conversion is also 2 By riding, it is possible to adapt the amplitude intensity distribution of the clutter to CA-CFAR.

(7) After conversion to uncorrelated white normal noise, the difference between the average of the time series data and the time series data after conversion is taken, and the sign of the data immediately before squaring is identified using the labeling by the sign function By selecting the positive and negative sign of the jump point from the CA-CFAR processing output, it is possible to switch the detection of the target of the low temperature or high temperature relative to the target and the ground surface corresponding to the change in the temperature of day and night. It is. That is, it is possible to switch between a target that is lower than the temperature and a target that is higher than the temperature. For example, a positive sign is selected during the daytime and a negative sign is selected during the nighttime.

(8) By logically calculating the value of the encoded output, both end points of the target signal rise and fall are detected, and the target existence section is automatically determined by considering the size of the target to be detected. Is possible.

(9) By using fractional differentiation, the original unimodal pulse is converted into a sign-inverted bimodal pulse, so that the target information contained in the middle part of the target signal is only the rise and fall of the target signal. By canceling each other and leaving a section where only the ground surface clutter exists, it is possible to apply the CA-CFAR process by eliminating the influence of mixed targets as much as possible from the statistics of the ground surface clutter.

(10) For time-series data determined not to have fractal properties, automatic target extraction can be supplemented by non-parametric CFAR processing.

  The present invention has been described above by taking the embodiment as an example. However, it is understood by those skilled in the art that various modifications can be made to each component and each processing process of the embodiment within the scope of the claims. By the way.

DESCRIPTION OF SYMBOLS 1 Far-infrared camera 2 Reference light source 3 Focus adjuster 4 Brightness adjuster 5 Attitude angle control device 6 Video monitor 7 Initial setting input device 8 Video recording device 9 Data conversion recording device 10 Window processor 11 1st target detection processing device 12 1st 2 target detection processing device 13 target display device 14 fractal property determination device 15 spectrum calculator 16 cutoff frequency detector 17 cutoff frequency adder 18 fractal dimension calculator 19 differentiator 20 inverse fractal filter 21 correlator 22 white tester 23 cutoff frequency Determinator 24 Latch 25 Fractional differentiator 26 Code converter 27 Chi-square distribution converter 28 Delayer 29 False alarm calculator 30 CA-CFAR processor 31 Binary converter 32 Target detection logic determination device 33 Target cell 34 Prefix Tapper group 35 Post-upper group 36 unit Function converter 37 Adder 38 Rank detector 39 Target detection determiner

Claims (8)

A far-infrared embedded automatic detection device that automatically detects a target embedded in the ground in real time by mounting a far-infrared camera on a traveling vehicle,
From the time-series data of one-dimensional luminance in the azimuth direction in the depression angle range limited by the far-infrared camera, it is complicated using the emissivity, radiation temperature and emissivity coefficient of all factors including the target and the ground clutter. without solving the heat conduction equation in the distribution to the boundary condition, we perform modeling by fractional Brownian motion, characterized in that to simulate the behavior on the emission intensity distribution of the far-infrared region of the surface clutter including vegetation far infrared Embedded automatic detection device.   The far-infrared buried object automatic detection apparatus according to claim 1, wherein a clutter suppression process in which a code discrimination function is incorporated into the CA-CFAR process is performed as a CFAR process adapted to the statistical properties of ground surface clutter in the far-infrared region.   By performing fractional differentiation on the one-dimensional luminance time-series data, the colored noise which is the ground surface clutter inherent in the far-infrared region is converted into uncorrelated white normal noise, and the output average of the time-series data after conversion is The far-infrared buried automatic detection apparatus according to claim 2, wherein the CA-CFAR processing is performed after taking a difference from the time-series data after conversion and squaring the result.   As a result, a negative jump point that becomes a negative value by taking the difference between the average of the time-series data converted to the uncorrelated white normal noise and the time-series data after conversion, and the absolute value is maximized. The far-infrared embedded automatic detection apparatus according to claim 3, wherein the amplitude intensity distribution of the clutter is adapted to the CA-CFAR processing by squaring.   By identifying the sign of the data immediately before squaring by taking the difference between the average of the time-series data converted to uncorrelated white normal noise and the time-series data after conversion, by using labeling with a sign function 4. The positive and negative signs of the jumping points are selected from the CA-CFAR processing output, and the detection of the relatively low or high temperature target is switched relative to the ground surface in response to changes in the temperature of day and night. Far-infrared buried object automatic detection device.   By identifying the sign of the data immediately before squaring by taking the difference between the average of the time-series data converted to uncorrelated white normal noise and the time-series data after conversion, by using labeling with a sign function , By selecting positive and negative signs of the jump point from the CA-CFAR processing output and performing logical operation on the encoded output value, both end points of the target signal rise and fall are detected, and the detection target 6. The far-infrared embedded automatic detection device according to claim 3, 4 or 5, wherein the target existence section is automatically determined in consideration of the target size.   By subjecting the one-dimensional luminance time-series data to fractional differentiation, the original monomodal pulse is converted into a sign-inverted bimodal pulse, so that the target information contained in the intermediate portion of the target signal is the target signal. CA-CFAR processing that reduces or eliminates the influence of mixed targets due to the statistical properties of the ground surface clutter by negating the appearance and leaving only the rising and falling points and becoming the section where only the ground surface clutter exists The far-infrared buried automatic detection device according to claim 2 which performs.   The far-infrared embedded automatic detection apparatus according to claim 1, wherein automatic target extraction is complemented by non-parametric CFAR processing for the one-dimensional luminance time-series data determined not to have fractal properties.